How to Start a Midjourney Side Hustle: Prompts, Pricing, and a Practical Playbook

If you're working full-time and can spare 5 to 10 hours a week, Midjourney becomes far more profitable when you stop treating it as a toy for random art and start targeting deliverables that sell -- YouTube thumbnails, social media assets, and ad creatives. Aiming for 10,000 to 50,000 yen (~$65-$330 USD) per month is realistic to start with. As of March 2026, the official Help Center often lists the Basic plan at roughly $10/month. Meanwhile, multiple user reports and explainer articles cite the Fast GPU Time add-on at about $4/hour, so treat those figures as ballpark and double-check the Help Center before committing. Whether a free trial is currently available also varies across sources, so confirm on the official page before building your payback model.

From hands-on experience, spending one hour on weekday evenings to draft three YouTube thumbnail concepts and refining them the next day is perfectly doable. V7's Draft Mode dramatically increases the number of rough concepts you can produce, making the draft-select-polish cycle much faster. This article walks through everything from choosing the right use case, designing prompts, building a production workflow, landing gigs, navigating commercial licensing and copyright, and handling taxes as a salaried worker -- ending with a concrete 7-day action plan.

What Actually Pays? Realistic Use Cases and Income for a Midjourney Side Hustle

High-Demand Use Cases

The sweet spot for Midjourney income isn't about making a single jaw-dropping image. It's about producing visuals that grab attention instantly and fit a specific purpose. While DALL·E 3 excels at faithfully following written instructions and generating variations, Midjourney's strength lies in aesthetic cohesion -- making it ideal for projects that need a consistent brand tone or visual world. Stable Diffusion offers deep customization and a steeper learning curve, but can deliver strong long-term ROI for those willing to invest.

YouTube and blog thumbnails are the most accessible starting point. YouTube thumbnails run 16:9 at 1280x720, so the practical workflow is generating the background or hero visual in Midjourney and adding text in Canva or Photoshop. Midjourney handles abstract mood requests especially well -- "business-themed but slightly futuristic" or "beauty niche with a luxury feel." When you need multiple compositional alternatives that closely match a text brief, DALL·E 3 can be the easier choice.

E-commerce product images and ad creatives are another strong fit. Think key visuals, mood-setting backgrounds, and seasonal campaign imagery rather than replacing product photography outright. For cosmetics, lifestyle goods, and apparel, Midjourney's texture rendering and color harmony pay off. When you need a "polished, premium world" for ad creatives, it typically gets you there faster than Stable Diffusion. That said, shots requiring exact product shapes are better handled with photography or a hybrid approach.

Social media content is also highly sellable. Instagram feed visuals, reel covers, and campaign graphics thrive on visual consistency -- exactly where Midjourney shines. In practice, a set of 9 social images at around 5,000 yen (~$33 USD) per set offers solid repeatability. Land two recurring clients and your monthly income starts to stabilize.

Blog hero images are beginner-friendly too. For outlets that want an atmospheric visual rather than an explanatory diagram, Midjourney is strong. Finance, career, AI, and lifestyle niches especially benefit, since stock photos alone tend to blend in. On the flip side, diagram-heavy publications or those needing precise text layouts are better served by DALL·E 3 or traditional design.

V7 has improved both quality and speed. Draft Mode is positioned as "10x faster than standard generation" per the official description. Multiple user reports indicate 4-image grids returning in roughly 10 seconds, though this is aggregated real-world data rather than an official spec. Keep the distinction between the official claim (10x faster) and user benchmarks (about 10 seconds) in mind.

Who This Is For

A Midjourney side hustle isn't just for people who can draw. It's for people who can articulate a visual world and iterate patiently. Prompts improve with specificity -- purpose, composition, texture, color, and use case all need gradual refinement. The person who picks one usable concept out of ten drafts consistently outperforms the person chasing a perfect first attempt.

Rule-followers thrive here. In commercial work, shortcuts like mimicking celebrities, replicating existing characters, or leaning too close to established brands cause the most damage down the line. Clients don't want a flashy one-off -- they want assets they can confidently publish every time.

Some basic design literacy also helps. For YouTube thumbnails, understanding eye flow, whitespace, text readability, and 16:9 framing matters more than the generation itself. The same applies to Instagram -- knowing how to center information within 1:1 or 4:5 dimensions noticeably improves deliverable quality. Midjourney produces strong raw material, but it won't make composition decisions for you. People who understand this distinction tend to grow fastest.

Conversely, if you need pixel-perfect control -- exact product shape reproduction, millimeter-level layout adjustments, or custom LoRA workflows -- Stable Diffusion is a better fit. If you want instruction-faithful variations in bulk, DALL·E 3 can be more efficient. Midjourney's sweet spot is thumbnails, ads, social media assets, and concept visuals -- work where polished atmosphere directly equals value.

The improved web interface has also widened the field. You're no longer tethered to Discord just to get started. Honestly, for a beginner looking to monetize quickly, learning Midjourney's sellable outputs first beats wrestling with Stable Diffusion's environment setup and extensions.

Income Ranges and How to Calculate Them

Skip the flashy success stories. The reliable formula is unit price x volume - fees and tool costs. A realistic initial target is 10,000 to 50,000 yen (~$65-$330 USD) per month. That's not a timid number -- it's the range where workload, skill building, revision cycles, and tool cost recovery don't create unsustainable pressure. Jumping straight to hundreds of thousands of yen invites a wall of sales and differentiation challenges before production skills even mature.

YouTube thumbnails offer a clear example. Based on observation of public listings, third-party aggregations typically place them at 500 to 2,000 yen (~$3-$13 USD) per piece, while gigs with quality requirements and recurring commitments can reach 1,500 to 3,000 yen (~$10-$20 USD) per piece. (Note: these ranges come from public listing observation, not official statistics.) Reference calculations (with the range I personally target alongside):

• Conservative range: 500-2,000 yen x 10-20 pieces = 5,000-40,000 yen/month (~$33-$265 USD)
• Practical target range: 1,500-3,000 yen x 10-20 pieces = 15,000-60,000 yen/month (~$100-$400 USD)

Budget 30 to 60 minutes per piece -- that includes not just generating the background, but text composition, export, and minor revisions.

Recurring social media deliveries also monetize well. A set of 9 images at 5,000 yen (~$33 USD), for example:

5,000 yen x 2 clients/month = 10,000 yen/month (~$65 USD)

Add spot thumbnail gigs and blog hero images, and you can reach the 20,000-40,000 yen (~$130-$265 USD) range. From experience, social media clients value consistent tone over high single-piece pricing. Small recurring work fits the side hustle model better than chasing big one-time payouts.

Tool cost recovery is straightforward. The Help Center commonly lists the Basic plan at about $10/month, and secondary sources report Fast GPU Time at roughly $4/hour (confirm the latest pricing on the official page, as figures may have shifted since writing).

From there, the monthly Basic fee is easy to recoup. Beyond that, staying within your base GPU allocation means profit accumulates quickly, and V7's Draft Mode for rapid concept generation aligns well with this payback math.

When using freelancing platforms like CrowdWorks and Lancers (Japanese platforms similar to Upwork and Fiverr) or skill marketplaces, your take-home drops. Coconala's service fee is 22% including tax, so a 10,000 yen (~$65 USD) listing nets about 7,800 yen (~$51 USD). CrowdWorks charges 20% on the portion of a contract under 100,000 yen, and Lancers charges 16.5%. The takeaway: price based on actual take-home, not the listed amount, or margins get thinner than expected. This is exactly why 10,000 to 50,000 yen/month is a sensible beginner target -- it's the range where you can turn a small profit while building your workflow and portfolio.

MidjourneyでGPU時間を無料でもらう方法（Fast HourボーナスをGET！）

Midjourney（ミッドジャーニー）でファストモードの残り時間を無料でGETする方法について紹介します！誰でも無料でGETできるお得な情報です！

blogcake.net

Getting Set Up: Plans, Web vs. Discord, and Essential Skills

Pricing Plans

As of March 2026, multiple explainer articles report that the general public free trial has ended, but the Help Center wording can change -- always verify the latest pricing page before signing up. As a minimum-cost benchmark, the Help Center frequently lists the Basic Plan at roughly $10/month. Treating that as a combined learning and prototyping expense makes the decision easier.

V7's Draft Mode pairs well with this initial investment. Generate a high volume of rough directions first, then only polish the winners -- this keeps GPU consumption down. The official description calls Draft Mode "10x faster than standard generation," and multiple real-world reviews report 4-image grids returning in roughly 10 seconds. Separating the official claim from user benchmarks helps readers avoid misunderstanding.

Web vs. Discord

Midjourney is now fully usable via the web interface, so the Discord-first assumption no longer applies. For beginners, the web version is the clear recommendation. It's easier to track generation history and selected images on screen, and operational mistakes are less frequent. In practice, the web UI also speeds up review cycles with clients, making it smoother to confirm which concepts to move forward with.

Discord still has its place. Once you're comfortable with commands and shortcuts, it offers a faster rhythm for consecutive generations and granular operations -- ideal for batch production. The practical sequence is start on web, add Discord as you scale.

In broad strokes, the web version is intuitive with low operational overhead, while Discord carries a slight learning curve but offers more flexibility for power users. The "feature gap" isn't critical for day-to-day side hustle work -- it surfaces more as you push deeper. Early on, choosing the environment where you can work without hesitation matters more than chasing feature parity.

One V7-specific note worth keeping in mind: multiple explainer articles and community reports mention that the personalization feature requires rating approximately 200 images on first use, but it's unclear whether the official documentation explicitly states "200." When onboarding, refer to the official help pages on Personalization and budget 30 to 60 minutes of initial setup time to be safe.

Skills and Environment

The skills you need aren't design-professional level expertise. They're the ability to reverse-engineer a goal into a structured prompt. Midjourney accepts Japanese prompts, but English tends to produce more consistent results. You don't need complex English, though -- short, separated terms for subject, composition, lighting, color, and mood already make a noticeable difference. Google Cloud's prompt design guide emphasizes goal clarity, specificity, and iterative refinement as fundamentals, and those principles translate directly to Midjourney.

For side hustle use, color basics and readability fundamentals matter as much as prompt skill. A stunning YouTube thumbnail background is worthless if the text disappears into it. On Instagram, simply understanding how to center information within 1:1 or 4:5 frames noticeably elevates deliverable quality. Midjourney strengthens your raw material, but presentation decisions remain yours.

On the environment side, you don't need a high-spec PC -- browser-based workflows are a genuine advantage. For client-ready output, pair Midjourney with an image editor for text overlay, cropping, and compression. Photoshop and Canva both work well here. Understanding the difference between PNG, JPEG, and WebP and knowing pixel dimensions by use case -- 1280x720 at 16:9 for YouTube, 1:1 or 4:5 for Instagram -- gives you a practical edge.

The people who grow fastest aren't the ones who generate the most images. They're the ones who can articulate delivery requirements upfront: "thumbnails optimized for click-through," "a soft palette targeting women without being saccharine," "e-commerce images with whitespace reserved for text." Clear purpose accelerates both prompting and image selection. While English prompts have an edge, Japanese-language Midjourney resources have expanded considerably, so the initial barrier is lower than it used to be. For a side hustle, the mindset of rapid prototyping and curating the best outputs beats deep tool mastery every time.

プロンプト戦略の概要  ｜  Generative AI on Vertex AI  ｜  Google Cloud Documentation

docs.cloud.google.com

Prompt Design That Sets You Apart: A 5-Element Template for Side Hustlers

Breaking Down the 5 Elements

When you use Midjourney for paid work, prompt quality directly determines deliverable quality. The framework that works is dividing every prompt into subject, use case, style, composition, and aspect ratio/parameters. Slotting these five elements explicitly beats writing a single intuitive sentence -- and makes revisions far easier.

In practice, I usually lock in where the image will be used, on what platform, and at what size before deciding what to depict. This ordering dramatically reduces revision requests. The single biggest factor is text-placement whitespace. A visual that looks beautiful on its own but collapses the moment a headline lands on it doesn't survive in client work.

Here's how the five elements break down:

Subject defines who or what you're depicting. "A woman" is too vague. "A businessperson in their 20s," "a cosmetics bottle with plain packaging," or "a laptop against a city night skyline" gives the generation a stable anchor. In client work, vague subjects breed misalignment. For people, specify age impression, professional context, and expression. For products, describe material and shape.

Use case is the differentiation core. You're not making "a beautiful image" -- you're making an image for a specific placement. YouTube thumbnails prioritize readability and whitespace. E-commerce product shots demand clean front-facing compositions on plain backgrounds. Instagram series need coherence when tiled. Google Cloud's prompt design guide emphasizes specifying purpose first, and this applies equally to image generation.

Style means texture and visual world. "Stylish" and "premium" are too subjective. Replace them with terms that evoke tangible visual qualities: matte finish, editorial photography, minimalist branding, high contrast, soft natural light. Weak style direction produces generically "AI-looking" images. Strong style direction transforms the same subject into client-ready output.

Composition covers viewpoint, distance, and placement. Front-facing or overhead? Bust shot or full body? Centered or offset? Without composition direction, you're leaving decisions to Midjourney. For thumbnails, "subject on left, empty space on right for text" makes post-production dramatically easier. For e-commerce, "front view, centered composition, plain background" eliminates ambiguity.

Aspect ratio and parameters are easy to underestimate but critical in practice. Aspect ratio maps directly to use case: 16:9 for YouTube thumbnails (YouTube Help recommends 16:9 at 1280x720), 1:1 or 4:5 for Instagram feeds, 9:16 for stories and reel covers. Adding quality, version, and Seed parameters strengthens reproducibility for series work and revisions.

In practice, this template structure works well:

Template format "Subject" + "Use case" + "Style" + "Composition" + "Aspect ratio/Parameters" + "Notes"

For example: "Businesswoman in her 20s, YouTube thumbnail, bold high-contrast editorial style, upper body positioned on left third, whitespace on right for text, 16:9." Notes might include "designed for text overlay," "simplified background," or "no brand logos." This structure lets you swap use cases and repurpose across platforms efficiently.

Replacing Vague Language and Designing Negatives

A common prompt failure is using abstract words verbatim. "Cool," "stylish," "cute," and "premium" feel convenient but actually introduce variance. The image in your head doesn't transmit at that resolution.

"A stylish banner" becomes far more directed as "minimal layout, matte texture, muted beige and charcoal, refined editorial look." "Cool" decomposes into bold typography, cinematic lighting, sharp shadows, high contrast. "Cute" maps to pastel tones, soft lighting, rounded shapes, playful composition. Converting emotional words into visual words is the single biggest lever for reproducibility.

Working backward from the use case changes which specific terms matter. YouTube thumbnails need to register at small sizes: high contrast, bold subject separation, clean background, space for headline. E-commerce product shots need: front-facing, symmetrical composition, plain white background, clean shadows, product-focused. Social media series need: consistent palette, signature accent color, repeatable composition.

Excluding unwanted elements matters for both quality and rights. Midjourney produces striking images but sometimes injects stray text, phantom logos, or excessive decoration. Negative prompt design addresses this. The approach is simple: don't just specify what you want -- explicitly state what you don't.

For side hustle work, organize exclusions along three lines:

1. Quality noise removal

Extra fingers, distorted hands, blurry details, messy background, low readability -- cut the elements most likely to break.

1. Production noise removal

Text, watermark, logo, frame, border -- exclude anything that doesn't belong in the deliverable. Text is better added in a design tool afterward.

1. Rights risk removal

Famous character style, specific brand logo, celebrity likeness -- steer away from associations that could create problems. Leaving this vague in side hustle work often produces assets that become unusable later.

Negatives don't need to be exhaustive -- cut what would cause problems for the specific gig. For thumbnails, target garbled text and overcrowded backgrounds. For e-commerce, target uncontrolled reflections and duplicate products. For social media, target inconsistencies in color tone across a series.

Using ChatGPT as a Prompt Co-Pilot

If prompt design is where you get stuck, ChatGPT works remarkably well as a verbalization assistant. Midjourney handles the image generation, but ChatGPT excels at structuring requirements and converting them into English-leaning instructions.

The workflow isn't complicated. Start by writing the project requirements in plain language. For example: "YouTube thumbnail. A woman in her 20s working from home. Whitespace on the right for a title. Bright but not childish. Blue-based. Clean feel." Then ask ChatGPT to break this down for Midjourney -- subject, use case, style, composition, and negative candidates come back organized.

The value here is translating the vague preferences in your head into production-ready language. When a client brief says "sort of elegant" or "feminine but not too sweet," feeding that directly into Midjourney is inefficient. Running it through ChatGPT first surfaces terms like soft neutral palette, clean editorial composition, subtle luxury, natural skin tones -- words that function as visual instructions.

Another practical use is building a personal vocabulary dictionary. As you take on more gigs, you'll notice recurring expressions. Map "clean feel" to clean light, minimal background, soft shadow. Map "premium look" to matte texture, deep contrast, premium packaging look. Having these ready means you stop rebuilding from scratch every time. In side hustle economics, reducing thinking time has a bigger impact on margins than reducing production time.

ChatGPT also semi-automates parameter selection. Ask it to organize recommended aspect ratios, composition approaches, and negative strategies for YouTube thumbnails, Instagram feeds, and e-commerce product shots, and you get a template draft almost instantly. Layer in your own frequently used style terms and you have a highly practical working blueprint.

This pairs well with the fact that the official description calls Draft Mode 10x faster than standard. User reports show 4-image grids returning in roughly 10 seconds, though exact timing varies by environment and load -- treat it as a guide. Rather than crafting one perfect English prompt, running multiple short iterations and refining only the winners improves working efficiency.

Before moving into use-case-specific examples, keep this template structure as your base: "Use case: ___, Subject: ___, Style: ___, Composition: ___, Aspect ratio/Parameters: ___, Negative: ___, Notes: ___" This format prevents information gaps across thumbnails, e-commerce, and social assets, and passes cleanly into ChatGPT.

Building Sellable Deliverables: Steps 1 Through 5

Step 1: Validate Demand

The first move isn't creating -- it's observing what's already selling. Skipping demand validation leads to images that look good but don't function as gig deliverables. Accessible starting points are CrowdWorks and Lancers (Japanese freelancing platforms similar to Upwork and Fiverr), Coconala, and social media job postings. Scan 10 recent listings for "thumbnail," "product image," and "social media assets." Thirty minutes is enough.

Focus on unit price, turnaround time, use case, revision count, and restrictions. Pricing reads differently depending on whether the gig is one-off or recurring. Turnaround phrasing like "next day" or "within 3 days" reveals urgency. Restrictions often contain implicit production rules: no famous character likenesses, no brand logo usage, no heavy imitation of existing ads, no thumbnails with insufficient text space.

Don't try to compute a precise market average at this stage. Public listing observation shows thumbnails clustering around 500 to 2,000 yen (~$3-$13 USD), while recurring or quality-gated gigs stretch higher. What matters is understanding what types of requests are active, under what conditions, and with what pain points. The consistent takeaway: clients want reliable, on-time, easy-to-swap practical assets -- not spectacular one-offs.

Keep notes simple. Five columns -- use case, target size/ratio, turnaround, restrictions, desired mood -- carry directly into the next step. YouTube defaults to 16:9 thumbnails at 1280x720. Instagram works with 1:1, 4:5, and 1.91:1. Locking in these specs early prevents the waste of regenerating a strong image in the wrong aspect ratio.

A common stumble is chasing keywords in listing titles. "Stylish," "premium," "high impact" are convenient labels but far too abstract on their own. The real information in a listing is who the audience is, where the image appears, what it needs to communicate, and what to avoid. Demand validation isn't just a sales exercise -- it's the foundation of your prompt design.

Step 2: Gather References

Once demand is visible, references sharpen your execution. The goal isn't imitation -- it's verbalizing the visual rules. Spend about 30 minutes collecting 5 to 10 examples from competitor accounts, established brands, and ads or posts in the same category.

Look at color, whitespace, text readability, and background treatment. YouTube thumbnails tend to simplify backgrounds heavily against strong subjects. E-commerce product shots keep subject contours clean with controlled reflections and shadows. Social assets prioritize consistency across a set over single-image impact. When analyzing references, always decompose what's working: Is it the cool blue tones? The generous whitespace creating a premium impression? The blurred background letting the subject pop? Putting it into words makes reproduction far more reliable.

Watch for over-attachment to a single reference. People who struggle with reference gathering tend to fixate on one favorite image, gravitating toward its composition and palette. The fix: never treat a single image as the answer. Line up 5 to 10 and separate commonalities from differences. Commonalities reveal the winning formula for that use case. Differences reveal where you can propose something fresh.

Check text readability at this stage too. Trying to finalize text inside Midjourney produces fragile results in practice, so plan the image with text placement space in mind. Thumbnails and banners especially suffer when backgrounds are packed with detail. Observing "where is the empty space?" during reference gathering prevents your collection from being a mood exercise with no production value.

Steps 3-5: Generate, Vary, and Prep for Delivery

For initial generation, don't aim for finished output. Use V7's Draft Mode to iterate rough and fast. The official description places Draft Mode at "10x faster than standard," and multiple user benchmarks report 4-image grids in roughly 10 seconds (environment-dependent, treat as a guide). In a 30-to-60-minute window, produce 20 to 40 concepts and save only the strong ones.

Separating "rough drafts via Draft Mode" from "polished finals" transformed hourly returns for me. Pouring time into refining a concept that turns out to be a dead end is the biggest efficiency killer. Reviewing a high volume of drafts first, then only investing in the viable directions, conserves focus for the finishing stage.

Next, lock the Seed and create variations. Focus on color swaps and layout alternatives -- three options tends to work well in practice. Think "one primary concept that fits the use case perfectly" and "two backup alternatives that are easy to swap in." Clients generally prefer choosing from a consistent direction rather than evaluating three wildly different ideas. Three variations with aligned intent outperform three with scattered directions.

A frequent mistake at this stage is sacrificing readability for visual impact. Thumbnails and social media announcement graphics especially suffer when background density overwhelms text. The fix is straightforward: ensure brightness contrast where text will land, simplify the area behind the subject, and check contrast first when creating color variations. A beautiful image and a usable image aren't always the same thing.

For finishing, use Enhance or upscale selectively. I typically limit these to surface-level refinement: final grain adjustment, noise reduction, and texture consistency for skin, metal, or fabric. Trying to fix composition at this stage often produces a different image entirely. Use enhancement to polish a strong draft, not to rescue a weak one.

Delivery prep means reviewing resolution, aspect ratio, whitespace, and text placement space. YouTube thumbnails need 1280x720 at 16:9 with a 2MB file size limit. Instagram content should work within 1:1, 4:5, or 9:16, with key elements centered for reuse flexibility. Format-wise: JPEG for photo-style images where file size matters, PNG for transparency or crisp elements, WebP for web-first delivery prioritizing lightness.

Rights issues are also easy to miss at the finish line. Details that weren't obvious during generation -- logo-like patterns, highly recognizable motifs -- can slip through. Before delivery, visually inspect the area around the subject for logos, symbols, or distinctive design elements. In side hustle work, the value isn't in generating the image -- it's in delivering something the client can use with confidence.

Use-Case Playbooks: Thumbnails, Product Shots, and Social Media Assets

When switching prompts by use case, start from "where will this image appear?" rather than "what should look good?" YouTube thumbnails need to communicate meaning in an instant. E-commerce product shots need the product front and center. Instagram series need visual coherence when tiled. Same tool, very different priorities.

For thumbnails, building contrast for eye guidance first changes performance. For e-commerce, specifying product contour and shadow carefully reduces revisions more than atmosphere does. Prioritizing "images the client won't struggle with downstream" over visual flash is what wins repeat side hustle work.

YouTube and Blog Thumbnail Prompts

Readability and contrast are the top priorities for YouTube and blog thumbnails. Can the subject be identified at small display sizes? Is there whitespace for text? Where is the person or motif placed? These determine the impression. Use 16:9 as the baseline, position the subject center or at a rule-of-thirds power point, and design for text overlay as a separate step -- reserving broad whitespace in the background area.

The most common thumbnail failure is an overly complex background that swallows the subject. For click-optimized images, reducing color count and creating a sharp light-dark split works better than adding detail. I often specify a brightly lit face or product against a slightly darkened background. This makes it harder for bold overlay text to get lost.

Key priorities: strong light-dark contrast, instant subject recognition, text whitespace, 16:9 screen design, single-frame emotional punch. Blog hero images follow a similar logic, but pulling back from YouTube's intensity level and prioritizing content alignment makes them more versatile.

Sample prompts:

dramatic YouTube thumbnail, Japanese businesswoman surprised expression, strong contrast lighting, dark blue background, clear negative space on the right for headline text, subject placed on left third, cinematic composition, bold visual hierarchy, 16:9 --ar 16:9 --stylize 150 --v 7 --no extra people, cluttered background, unreadable text, watermark, logo

blog eyecatch image, minimalist laptop workspace, bright subject and darker background for eye guidance, clean composition, center-right empty space for title, high readability, modern editorial style, white and navy palette, 16:9 --ar 16:9 --v 7 --no messy desk, too many objects, text, icons, low contrast

YouTube thumbnail for finance topic, confident Japanese man in suit, sharp rim light, red and black high contrast background, subject centered, space above and right for large Japanese text, attention-grabbing, polished, realistic --ar 16:9 --stylize 200 --v 7 --no distorted hands, duplicate face, crowded scene, small objects, watermark

For this use case, writing "who goes where" and "where text goes" as a pair raises the hit rate. When prioritizing readability, high readability, strong contrast, and negative space for headline text are particularly effective terms.

E-Commerce Product and Ad Creative Prompts

E-commerce and ad imagery differs from thumbnails in that the subject must be unambiguous, the spec must be communicable, and the asset must be easy to swap. Aspect ratios of 1:1 or 4:5 work well, backgrounds should lean plain or monochrome, and compositions should make the product's contour cleanly readable. Generous whitespace makes it easier for the sales page or ad platform to layer in copy.

Here, product shape, material texture, and shadow treatment take priority over mood. Ambiguous contours and unnatural contact shadows tend to trigger revision requests. Specifying "clean silhouette," "soft shadow on surface," and "accurate product edges" lifts product-shot quality. Even when creating a more atmospheric ad visual, the product standing clearly comes first.

Key priorities: subject dominance, plain or monochrome background, natural contact shadow, generous whitespace, 1:1 or 4:5 ratio, minimal color drift. For ad banner assets, avoid centering the product too aggressively -- leaving copy space on one side makes downstream production smoother.

Sample prompts:

premium skincare bottle product shot, centered hero product, clean white seamless background, accurate bottle silhouette, soft natural shadow on surface, minimal luxury lighting, generous negative space, ecommerce ready, square composition --ar 1:1 --v 7 --stylize 80 --no text, label distortion, extra products, busy background, watermark

wireless earbuds advertising image, matte black case, single-color light gray background, crisp product edges, realistic reflection, soft drop shadow, subject dominant, space at top for copy, modern commercial photography, 4:5 --ar 4:5 --v 7 --no hand model, extra accessories, clutter, typography, logo

organic coffee package hero image, product as main subject, warm beige solid background, visible front-facing packaging, clean contour, subtle shadow, premium ad visual, large empty area on upper left for campaign text, polished studio look --ar 4:5 --stylize 100 --v 7 --no messy props, duplicate package, unreadable package details, watermark

For e-commerce, over-styled backgrounds make product pages look cluttered in grid view. Even for ad creatives, separating the atmosphere shot from the product-hero shot is more practical. Secure the standalone product cut first, then build atmosphere shots -- this sequence makes gig delivery smoother.

Instagram Asset Prompts

Instagram content thrives on series-level coherence over single-image impact. For feed posts, 1:1 or 4:5 is the standard. Fixing brand colors, whitespace rules, borders, light direction, and texture tone across multiple images prevents the visual world from fracturing. Nine images tiled together with consistent rhythm matters more than any one standout.

For recurring account management, defining "this account uses these colors, this whitespace, this mood" upfront beats rebuilding from scratch every time. For example: only beige and deep green, subject centered, equal top and bottom margins, thin border frame. Midjourney naturally produces single-image beauty, but left unguided, each generation drifts slightly. For series work, loading more fixed conditions into the prompt is essential.

Key priorities: locked color palette, unified style, explicit border and whitespace rules, brand world, 1:1 or 4:5 reproducibility. For template-style production, center the subject, add whitespace above and below, and keep the background light with low information density.

Sample prompts:

Instagram post series for wellness brand, consistent beige and sage green color palette, minimalist editorial style, centered subject, soft natural light, clean margins, subtle thin border frame, calm premium mood, square feed design --ar 1:1 --v 7 --stylize 120 --no random colors, busy background, text, watermark, inconsistent lighting

Instagram carousel cover, lifestyle beauty theme, warm neutral palette, elegant composition, large top and bottom whitespace for future typography, unified visual language, soft shadows, premium social media asset, 4:5 --ar 4:5 --v 7 --no neon colors, clutter, extra props, logo, chaotic layout

Instagram content series for café brand, fixed terracotta and cream palette, handcrafted cozy mood, consistent camera angle, centered hero item, clean frame, repeatable template feeling, polished branding visual, 4:5 --ar 4:5 --stylize 90 --v 7 --no mixed styles, dark moody lighting, crowded scene, text artifacts

The overlooked failure mode for Instagram is assets that look great individually but lose all coherence when tiled. Prevention requires embedding the same core terms in every prompt. consistent color palette, unified visual language, and repeatable template feeling are useful phrasing for series cohesion. For side hustle accounts where preserving the visual world is the deliverable, the ability to intentionally produce "slightly similar every time" directly affects client retention.

Midjourney vs. DALL·E 3 vs. Stable Diffusion: Choosing the Right Tool for Side Hustle Work

Three-Tool Comparison

Choosing an AI image generator for side hustle work isn't about raw superiority. It's about where each tool saves you time and where it lets you charge more. Midjourney, DALL·E 3, and Stable Diffusion have distinctly different strengths. Midjourney wins on single-image polish, DALL·E 3 wins on instruction-following speed, and Stable Diffusion wins as you invest more into customization.

Midjourney's defining advantage is the probability that any given output "looks professional" on first glance. For thumbnails, ads, and social visuals, the way light wraps, textures render, and tones harmonize produces deliverables that pass client review more readily. V7's Draft Mode further supports side hustle workflows by enabling rapid concept iteration -- 4-image grids returning in roughly 10 seconds makes directional exploration highly efficient.

DALL·E 3 shines when you need to translate a client's verbal instructions directly into an image. "Empty the left side of the desk," "shift the person slightly right," "give me three options on the same theme" -- these micro-adjustments land faster. In side hustle work, reducing alignment back-and-forth during the early meeting stage often matters more than polish. I sometimes lock composition and whitespace balance using DALL·E 3 first, then layer in lighting and texture through Midjourney for a final that's both aligned and visually strong.

Stable Diffusion isn't a "quick first sale" tool -- it's a tool that grows stronger as you shape it around specific gig requirements. LoRA and ControlNet unlock deep control over specific art styles, poses, compositions, and product presentation rules. The learning investment is real, but for niche gigs and reproducibility-critical recurring work, long-term ROI emerges. It becomes especially valuable once you hit the stage where "my output looks the same as everyone else's" starts bothering you.

Recommendations by Use Case

Choose based on the work you want to take, not the images you want to make. Whether you're batch-producing thumbnails, polishing a single ad hero, or building a distinctive style for recurring clients determines which tool fits best.

For YouTube thumbnails, social media visuals, and ad creatives -- work where first-impression impact is the primary metric -- Midjourney is a strong default. Background atmosphere and subject presence come together quickly, and visual world consistency supports recurring gig retention. Clients in these categories value "that brand-consistent look" more than technical novelty.

For gigs demanding instruction-faithful images delivered fast, DALL·E 3 fits naturally. Banner size variants, same-theme color swaps, background alternatives, whitespace adjustments -- its high fidelity translates directly into time savings. In side hustle economics, reducing the number of alignment round-trips with clients often matters more than crafting one exquisite piece.

For gigs requiring custom models or granular control, Stable Diffusion is the answer. Specific art style reproduction, pose locking, composition locking, and sketch-to-final precision are areas where Midjourney and DALL·E 3 run out of runway. Mastering Stable Diffusion takes longer, but the gap between you and competitors becomes a durable technical asset. It pays off most during the phase where you're gradually raising your rates.

A practical summary: Midjourney gets you to sellable output fastest, DALL·E 3 gets you to aligned output fastest, Stable Diffusion gets you to uniquely differentiated output over time. Beginners monetize most quickly with Midjourney, but DALL·E 3 can be less draining for revision-heavy work. Stable Diffusion is best viewed as a strategic investment for building advantage in niche or recurring markets.

Maximizing ROI Through Tool Combination

In practice, splitting responsibilities across tools by production phase yields better ROI than forcing one tool to do everything. Side hustle hours are limited, so designing workflows that minimize "where do I get stuck?" time pays off.

The combination that consistently works well: DALL·E 3 for composition lock-in, Midjourney for final visual polish, Stable Diffusion for precision gigs requiring tight control. DALL·E 3 settles layout direction and whitespace balance quickly, making early client alignment faster. Midjourney then layers on lighting, texture, and atmosphere -- transforming the draft into a professional deliverable. For gigs with specific style or shape constraints, Stable Diffusion reduces revision loops.

This sequence works because each tool's strengths cleanly complement the others' gaps. Midjourney produces beautiful output, but trying to front-load every specific constraint into it can slow down directional changes. DALL·E 3 is better suited for the "what goes where" phase. Stable Diffusion handles "this gig's unique reproducibility requirements" at the final layer.

In side hustle work, reducing revision count matters more than any single tool's output quality. Rather than aiming for perfection on the first generation, using DALL·E 3 to minimize misalignment, Midjourney to elevate visual value, and Stable Diffusion for precision-only gigs shortens time to delivery while raising deliverable quality. For ad creatives and social media management assets especially, separating "fast draft" from "polished final" across tools significantly reduces production stress.

If you had to summarize when Midjourney is the right call in one sentence: when visual credibility directly drives win rates and client retention. Thumbnails, ads, and social visuals -- deliverables judged on first impression -- are where Midjourney's quality and aesthetic consistency become a weapon. Layer in DALL·E 3 and Stable Diffusion as needed, and you move from "someone who can generate AI images" to "someone who delivers sellable visuals efficiently."

Finding Gigs, Listing Services, and Setting Prices

Platform Comparison

Gig acquisition falls into three main channels: task-based platforms like CrowdWorks and Lancers (Japanese freelancing platforms similar to Upwork and Fiverr), fixed-price listing platforms like Coconala, and direct commissions via social media. The question isn't which is best -- it's which fits your current stage.

CrowdWorks' official worker fee page shows 20% on the portion under 100,000 yen (~$660 USD), 10% on the 100,001-200,000 yen portion, and 5% above 200,000 yen. Lancers' official FAQ lists a 16.5% service fee. Lancers is lighter on fees, but the real differentiator in task-based platforms is your ability to win proposals against the listing's requirements. Reading the use case, delivery format, and revision scope first and tailoring your pitch accordingly builds traction. YouTube thumbnails and social media assets are particularly approachable because the listing text usually makes the pain point visible -- giving even newcomers a concrete angle.

Coconala works differently -- you list a service and wait. Coconala's official help page states a 22% service fee including tax, so a listed price translates to a thinner take-home. The upside is strong product framing: "YouTube thumbnail specialist," "Instagram 9-image set," "soft-tone visuals for women's brands." If task-based platforms lean toward sales, fixed-price listing platforms lean toward product design. Even with a thin track record, a well-packaged listing can serve as an effective entry point.

Direct commissions via social media carry no platform fees, but trust-building works differently. Rather than selling immediately, you need to demonstrate that "this person can be trusted to deliver." Posting use-case-specific portfolio pieces on Instagram or X shifts the response rate. For thumbnails, hero images, and social banners especially, clients evaluate "can this person consistently produce this style?" over single-image brilliance.

For profiles and portfolios, curate about three pieces per use case. YouTube thumbnails at 16:9, blog hero images, Instagram feed assets at 1:1 or 4:5 -- segment by platform. Consistency and explicit delivery specs matter more than volume. Noting "1280x720 at 16:9" for YouTube or "1080px base" for Instagram signals practical experience. Adding a simple line like "AI-generated; rights-verified" reduces client anxiety significantly. This matters more than most people realize -- the deciding factor is often "can I use this safely?" rather than "does this look amazing?"

Midjourney's prototyping speed also functions as a sales tool. V7's Draft Mode, officially described as 10x faster, means directional concept drafts return quickly -- useful for pitch preparation. In task-based bids, it powers fast initial roughs. For fixed-price listings, it enables rapid sample production. On social media, it supports consistent visual posting cadence.

Pricing Table

"Public observation" reflects ranges observed across public freelance listings. The "practical target" is the range I find achievable for quality-gated and recurring work.

The logic behind this structure goes beyond market rates -- it makes the work cleanly divisible. "Present 3 concepts, deliver the selected 1, include 1 revision" stabilized client satisfaction noticeably. Clients get the comfort of comparison, and you get clear boundaries on what's covered. Bonus: clients sometimes want to use a non-selected concept for another purpose, which opens a natural add-on pricing conversation.

Timing price increases on a clear milestone rather than gut feeling prevents hesitation. Three completed gigs in the same use case is the recommended trigger. By that point, you've identified the recurring friction points in composition, text whitespace, and delivery format. A 20-30% increase at that stage is absorbable without a steep drop in win rate. For example, a 2,000 yen thumbnail becomes 2,400-2,600 yen; a 6,000 yen social media set becomes 7,200-7,800 yen.

Factor in take-home. Coconala's 22% fee means a 10,000 yen listing nets 7,800 yen (~$51 USD). When building your pricing table, evaluate "hours of work per amount received" rather than the listed price.

Tool cost recovery is also worth quantifying. Public sources and explainer articles commonly cite the Basic plan at about $10/month, with Fast GPU Time add-ons appearing at roughly $4/hour in secondary reports (always confirm current pricing on the official Help Center). Producing 20 pieces per month at 2,000 yen (~$13 USD) each generates 40,000 yen (~$265 USD) in gross revenue. Subtract tool costs and fees from there to assess viability.

Proposal Templates and Tips for Winning Gigs

Proposals that fit in one scroll outperform lengthy essays. On CrowdWorks and Lancers, clients compare multiple proposals quickly, so reducing reading effort raises your pass rate. Structure proposals as: use case understanding, production process, turnaround, revision count, restricted content compliance, and a quick rough draft.

A practical flow:

"I see this is for a business-category YouTube thumbnail. I'll design with CTR in mind, prioritizing eye flow toward the subject and reserving whitespace for headline text. I'll present 3 directionally distinct concepts, polish the selected one, and deliver. First draft by [date], with 1 revision included in the base price. I avoid celebrity likeness, character imitation, and brand-adjacent expressions, producing within rights-verified AI asset boundaries. I've attached a quick rough in a similar tone for reference."

The strength of this structure is that it communicates "I understand the job" before claiming any skill. Google Cloud's Vertex AI documentation emphasizes clarifying purpose, adding specificity, providing examples, and iterating -- proposal writing follows the same logic. What you'll make comes first; how impressive you are comes second.

The simplest path to higher win rates is replacing abstract claims with concrete specifications. "High-quality work" loses to "16:9 at 1280x720 with reserved headline whitespace." "I can do social media" loses to "1:1 and 4:5 Instagram assets with consistent tone." Embedding platform specs into the conversation shifts perception from "someone who generates AI images" to "someone who builds for the deployment context."

Attaching a single quick rough with the proposal is a tactic that works well. Not a finished freebie -- a directional draft for alignment. Draft Mode makes this low-cost; concept images return in seconds, keeping proposal prep lightweight. Including a rough makes client replies more specific and reduces alignment gaps.

Conversely, proposals that say "I can do anything" or "I'll do it cheap" underperform. Low-price gigs erode profit the moment revisions increase. Short, structured proposals covering use case understanding, process, and compliance consistently convert into recurring relationships. For AI image gigs specifically, proactively stating restricted content boundaries -- no famous character imitation, no celebrity likeness, no brand-adjacent expressions -- raises trust in a way that's subtle but effective for long-term retention.

Common Mistakes and How to Avoid Them

Rights and Legal Issues

The most dangerous failure in AI image side hustle work isn't visual quality -- it's getting stuck on rights. The most frequent scenario: producing output that leans too close to a famous character or celebrity. When a client brief says "something like [character name]" or "in the style of [brand]," the temptation to match is strong, but accepting that direction uncritically often produces images that become unusable after delivery. In practice, it's not just the name that matters -- when hairstyle, outfit, color scheme, and signature pose align, the resemblance becomes legally fragile.

The effective countermeasure is locking restricted territory at the prompt stage. Beyond avoiding celebrity names and franchise titles, skip phrasing like "in the style of" that implies derivation. Add distinctive features of known properties to the negative prompt. Then reframe with original composition and abstracted motifs. "The fierce expression of a popular anime protagonist" becomes "backlit, low angle, sharp gaze, warm energy tones." The shift: from imitation work to usable work.

Another frequently overlooked issue is proceeding with vague usage terms after delivery. Platforms haven't uniformly established granular standards for AI-generated content, making per-gig agreements the de facto rules. Clarifying "what is this image for?", "where will it be published?", and "are you expecting likeness to any existing IP?" as production conditions upfront significantly reduces post-delivery disputes. The legal landscape may seem complicated, but on the ground, "don't create risky resemblances from the start" is the single most effective rule.

Quality and Readability

A beautiful generated image that can't support readable text is weak as a commercial product. The common thumbnail and ad creative failure: an over-detailed background that leaves no room for text overlay. Midjourney's output density runs high by default, so without whitespace direction, both subject and background compete for attention. The result: a thumbnail where information doesn't register, regardless of visual quality.

Prevention means designing text space before image generation. I routinely specify "whitespace for 2-3 lines of heading in the upper right" or "the left third should have a simple background." Working within platform specs -- 1280x720 at 16:9 for YouTube, for example -- and keeping key elements away from the edges stabilizes downstream layout. Backgrounds built with color fields and lighting rather than dense props are dramatically easier to layer text onto.

Overusing vague terms also degrades quality. "Nice," "stylish," and "premium" without decomposition produce wide output variance. Google Cloud's Vertex AI documentation reinforces that prompts benefit from clear purpose, specificity, and iterative refinement -- this applies identically to image generation. I keep a semi-fixed reference vocabulary: "premium" decomposes into material texture, light direction, color temperature, and background treatment. Running abstract terms through ChatGPT to convert them into concrete visual language improves output reproducibility.

One practice that made a measurable difference: making scaled-down review a mandatory pre-delivery step. After establishing "if text isn't readable at reduced size, it doesn't ship," revision requests dropped by roughly 30%. Images that look polished at full resolution can fall apart at thumbnail or mobile display sizes. Checking headline readability and subject visibility at reduced scale produces much more realistic quality assessments.

Cost and Time Management

The side hustle profit killer is pursuing perfection on a single image and burning through GPU time and working hours. Public sources report Fast GPU Time add-ons at roughly $4/hour in secondary reporting, but always verify the latest pricing on the official page. Unplanned generation runs thin margins, so managing GPU spend through the Draft-to-finish workflow is essential.

Regenerating from scratch without locking Seeds also quietly inflates costs. When a strong composition appears, use it as a base for adjustments -- this makes same-direction comparisons far easier. In side hustle work, "how reproducibly can you hit a consistent direction?" matters more than "how often can you roll a stunning one-off?" Seed locking isn't flashy, but it directly speeds up revision response.

Delivery format mismatches are another common time drain. The image itself might be strong, but misaligned aspect ratio, resolution, file extension, or color space triggers re-exports or re-generation. YouTube requires 16:9 at 1280x720 within 2MB. Instagram renders differently across 1:1, 4:5, and 9:16. Capturing platform and spec requirements at the time of accepting a gig and locking format upfront prevents late-stage rework. Honestly, more working time is lost to this kind of specification gap than to production difficulty.

For time management, estimating per-gig effort across "prototyping," "polishing the selected concept," and "export" as separate phases keeps things stable. The 3-concept-present-1-deliver structure mentioned earlier pairs well with cost control. Draft a batch of concepts quickly, invest GPU time and focus only on the concepts that gain traction. This workflow makes the ceiling visible and prevents the trap of accepting low-price work that consumes disproportionate effort.

Commercial Use, Copyright, and Tax Considerations for Salaried Workers

Commercial Use: Permissions and Limits

Midjourney allows commercial use under paid plans, but "permitted by terms of service" and "fully legally secure" are separate concepts. The platform's usage license and questions around whether copyright attaches to AI-generated images or whether outputs might infringe third-party rights are distinct issues. Especially given that legal interpretation of AI-generated content remains in flux across jurisdictions including Japan, relying solely on platform terms for confidence in client work carries risk.

The safest lane for side hustle work: abstract backgrounds, original concept visuals, and imagery that doesn't directly identify specific products or individuals. The lane to avoid: output mimicking famous characters, creatives heavily echoing a specific brand's visual world, or visuals closely resembling real public figures. These deliver high initial impact but frequently trigger post-delivery takedowns or swaps, costing more in time and trust than the gig was worth. When a client casually asks for "something like [brand/character]," the provider needs to draw the line first -- otherwise accidents happen.

I include AI disclosure, restricted content boundaries, and reuse scope in every order sheet from the start. For example: "No imitation of famous characters, celebrities, or existing brands. Reuse scope defined per project." Since adopting this practice, disputes over "the style wasn't what I expected" and "I want to use this exclusively across other projects" have dropped significantly. Defining the scope of production work upfront matters more in practice than debating the commercial use license itself.

Copyright and Derivation Risk Management

The most friction-prone area in generated imagery isn't copyright in the abstract -- it's perceived derivation. Even when you created something from scratch, if it looks close enough to an existing work to suggest it was based on one, explaining becomes difficult. When a famous character's outfit, distinctive color scheme, signature composition, and logo-adjacent design elements align, "it's coincidental" doesn't hold up well. Even without tracing, leaning too heavily on a reference image crosses into practically dangerous territory.

The legal standing of AI-generated imagery still has unresolved dimensions: how much originality is recognized, what level of human creative involvement is required, and how derivation is assessed when similar images appear are all questions where the answer varies by case. For client work, this makes it practical to agree upfront on "how far can the deliverable be reused?", "is exclusivity expected?", and "what happens if a swap is needed?" I find keeping usage scope, exclusivity, and swap handling simple and explicit works better than overly legalistic language.

Other people's images and prompts also require care. Referencing publicly shared work is different from unauthorized reproduction. Uploading someone's image to generate derivatives, copying a commercially listed prompt nearly verbatim, or mass-producing from another creator's portfolio images are all high-risk. "AI processing makes it different" doesn't hold as a defense. In client work especially, discovery of source material destroys trust faster than it destroys deliverables.

Tax Filing and Employment Policy Checks for Salaried Workers

For salaried workers taking on AI image side hustles, employment policies matter as much as copyright. Whether your company requires approval, notification, or places restrictions on the type of side work determines your operating space. An overlooked risk: if your employer's definition of work product or intellectual property output is broad, deliverables that appear continuous with your day job can create friction. Design, marketing, and creative department employees are especially vulnerable to blurred boundaries -- consciously separating your employer's work from side hustle work across subject matter, hours, and equipment is essential.

Using company computers, employer-licensed Adobe products, workplace-sourced assets, or client information obtained through your job for side hustle work should be avoided. This isn't just etiquette -- it's a confidentiality and asset misuse issue. Even casually using Midjourney on a company PC during breaks creates explanations you'd rather not have to give. Separate your side hustle production environment entirely: accounts, devices, and file storage.

On taxes, the general framework in Japan is that salaried workers must file a tax return when side income exceeds 200,000 yen (~$1,320 USD) annually. Note that this threshold applies to income (revenue minus expenses), not gross revenue. Tool subscriptions and necessary expenses are deductible. AI image side hustle amounts per gig are small, but they accumulate faster than expected -- a steady stream of thumbnail and social media deliveries can reach the threshold well before year-end.

Resident tax handling is another detail worth noting. Many salaried workers prefer their employer not to know about side income, and the typical discussion centers on whether resident tax can be filed via direct payment (ordinary collection) rather than payroll deduction. However, actual processing depends partly on the municipality, so looking beyond "whether to file" to "how resident tax gets processed" is important for avoiding unintended disclosure. Tax administration is tedious when deferred, so tracking revenue and expenses separately alongside gig management keeps things manageable.

Note: The tax details above reflect Japan's tax system. If you're based outside Japan, check your local tax filing thresholds and obligations for side income.

Your First 7-Day Action Plan

The Week's Goal

This week's target isn't just trying Midjourney -- it's picking one use case, producing a showable portfolio piece, and taking action toward your first sale. Choose exactly one: thumbnails, product images, or social media assets. Fixing the use case makes prompt improvement visible, and portfolio and proposal creation become straightforward.

Day 1: Register for Midjourney Basic, verify current pricing on the Help Center, and choose your single use case. Trying everything at once scatters your portfolio direction. For the first week, "someone who looks reliable for this specific use case" beats "someone who can do anything."

Days 2-3: Build 3 template prompts dedicated to your chosen use case. English fluency isn't required -- write requirements in your native language, run them through ChatGPT for English conversion, and push toward concrete visual terms. Decompose into subject, composition, lighting, color, and text-overlay whitespace for reproducibility. Simultaneously, collect 5 reference images and verbalize the mood -- this speeds up revisions later.

Day 4: Use V7 Draft Mode to produce 20 concepts in one session. Draft Mode is built for this kind of iteration, with 4-image grids returning in roughly 10 seconds. Don't polish all 20 -- advance only the top 3 to finishing. Export at the correct aspect ratio for your use case. Thumbnails at 16:9, Instagram feed assets with centered composition for portrait orientation. Matching the output format from the start adds practical credibility.

Day 5: Turn your best piece into a portfolio entry. If you can show 3 pieces organized by use case, even better. But don't just post the image -- briefly note the intent behind it, the prompt design approach, and that you avoided restricted content territory. Trust signals outweigh visual impressiveness here.

Day 6: Check demand on CrowdWorks, Lancers, or Coconala (or equivalent platforms like Upwork, Fiverr, or Etsy if you're outside Japan) and either submit at least one proposal or list one service. Start at the lower end of your range and watch what generates response. At this stage, understanding what resonates matters more than pricing aggressively. Reading listing descriptions makes one thing clear: clients want functional images for specific placements, not fine art.

Day 7: Review the responses from proposals or listings and update your proposal template. If traction was low, the issue may not be the image itself -- phrasing of the use case, delivery preview, and revision handling often matter more. Adjusting one element per week sustains improvement better than periodic overhauls.

What to Do in Weeks 2 and 3

For the next two weeks, resist the urge to expand into new use cases. Stay on the same one and refine. If you chose thumbnails, keep making thumbnails. Switching too early dilutes the feedback signal. In early side hustle stages, reproducibility creates more value than breadth.

Start by taking the best-performing template prompt from week 1 and iterating an improved version. Adjust word order, composition direction, color terms, and whitespace handling incrementally. Google Cloud's Vertex AI prompt design principles -- clarify purpose, add specificity, iterate through testing -- apply directly to Midjourney side hustle work.

On the sales side, target 3 proposals in the same use case the following week, or post a promotional piece on social media to gauge response. Don't rewrite proposals from scratch each time -- template "what I make," "how I deliver," and "what I cover" and only adjust per-gig details. Sustaining both production and sales requires writing efficiency.

For listing-based platforms, a Coconala service with a use-case-specific title, sample images, and a visible production process works well. For proposal-based platforms, read CrowdWorks listings for signals: is the client in a hurry, looking for recurring work, or expecting heavy revisions? Adapt the proposal accordingly. At zero-track-record stage, "easy to work with" outperforms "talented" as a selling proposition on both channels.

The deliverable from these two weeks isn't a big revenue number. It's a solidified three-piece kit: template prompts, portfolio samples, and a proposal template. Once those are locked in, operating within 5 to 10 hours per week becomes sustainable, and per-gig hesitation drops. Side hustles grow faster from steady incremental improvement than from bursts of intensity.