- Home
- Skills
- Image & Video
- Image Studio: Cheap Drafts, Premium Finals
Works with the AI Agents you already use
Image Studio: Cheap Drafts, Premium Finals
Premium image models cost 10–50× more than cheap ones, and most of the money goes on prompts that weren't right yet.
$5
Image Studio: Cheap Drafts, Premium Finals
Example session with this skill installed
Make a 16:9 hero banner for my coffee subscription site "Slow Pour". Flat illustration, warm morning colors, a pour-over kettle and a cup. The only text should be the headline "Slow Pour". Try a few cheap models first.
- Read your context and instructions
- Compiled the image studio
- Generated the image
Fan-out plan
Headline only, so I'm using cheap models tagged text-head, plus two that are strong on clean flat graphics. Same prompt for all of them:
Flat graphic artwork, 16:9. A pour-over kettle pouring into a ceramic cup, warm morning light, soft peach and cream palette, generous empty space on the left. The ONLY text anywhere in the image is the two words "Slow Pour", large, on the left.
| Model | Why it's in | Quote |
|---|---|---|
| flux-schnell | fastest, good on short headlines | $0.0039 |
| minimax-image-01 | cheap, different style | $0.0046 |
| gpt-image-25-flare (medium) | best composition in our tests | $0.0156 |
| gpt-image-2 (low) | reliable text | $0.0156 |
| qwen-image-fast | precise text rendering | $0.039 |
Total: $0.079. Under the $0.50 cheap-tier limit, so running now.
Results
Saved to ./images/2026-11-12/:
| File | What came out |
|---|---|
20261112-101530-slowpour_hero_flux-schnell_01.png | Headline perfect, kettle simple, a bit flat |
20261112-101530-slowpour_hero_minimax-image-01_02.png | Nice palette, but the font came out handwritten |
20261112-101530-slowpour_hero_gpt-image-25-flare_03.png | Best scene: steam, light, empty space where asked |
20261112-101530-slowpour_hero_gpt-image-2_04.png | Clean type, cup slightly cropped |
20261112-101530-slowpour_hero_qwen-image-fast_05.png | Crisp text, busier background |
Each file has a .json log next to it with the exact prompt, parameters and actual cost. Actual total: $0.079.
Final
Which one should I finish on?
Auto: I'd take the flare composition to gpt-image-2 at high quality for crisp type (high quality is priced per run; I quote it first)
2. nano-banana-2 at 2K, $0.065: sharp, large, good with text
3. qwen-image, $0.052: the cheapest premium option with precise text
4. nano-banana-pro at 4K, $0.195: maximum detail
Say which, and I'll quote it exactly before running.
20261112-101530-slowpour_hero_flux-schnell_01.png
PNG · 1536×1024
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Premium image models cost 10–50× more than cheap ones, and most of the money goes on prompts that weren't right yet. This skill makes your agent work like a careful designer: try the idea cheaply on several models, pick the composition you like, then pay for one premium render.
What it does
- Fan-out. One prompt, identical for every model, across 3–6 cheap models (from $0.004 per image). You see side by side which model got the layout, the style and the text right.
- Routing by tags, not guesswork. Every model carries tags from real runs:
text-head(reliable on a short headline),text-small(also captions and buttons),scene(keeps many objects in place),strict(adds nothing you didn't ask for),4k,refs. The agent picks models that fit the job. - Premium final. After the fan-out the agent offers Auto plus the three best premium models for the task (GPT Image 2 high, Nano Banana 2 / Pro, Alibaba 4K), each with its price and why.
- No surprise bills. Every run is quoted first. Cheap fan-outs under $0.50 just run; anything premium waits for your explicit yes. One yes covers one run.
- Files you can find later. Images go to
./images/YYYY-MM-DD/with one batch prefix per run, plus a JSON log next to each file with the exact prompt, parameters and actual cost.
Field notes included
The skill ships with what we learned on fan-outs of 8–15 models on real banners and illustrations, so your agent doesn't relearn it on your balance:
- Short headlines work almost everywhere; small text separates the models.
- "Banner" makes some models draw a whole landing page. The fix that works: The ONLY text anywhere in the image is "…".
- Negative instructions are unreliable on every vendor: "no spiders" was ignored by 4 of 15 models.
- Kerning and hex colors can't be prompted; letter logos belong in SVG.
- Which cheap model is best for busy explainer scenes, which is the most disciplined on minimal graphics, which one confuses semantics.
Where it stops
- Text-to-image and reference-based generation only; no video, no upscaling, no background removal.
- Logos, faces and brand assets need a real reference file. The agent won't describe your logo in words.
- Non-Latin text (Cyrillic, CJK) is not yet verified on every model; the skill tells the agent to fan out and check by eye.
No API keys or provider accounts: every model runs as an Agensi tool, paid per image from your balance. Failed runs are free.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
Cannot guarantee hex color accuracy or kerning. Negative prompts are unreliable across all models. No logo design from text; requires reference files. Requires Agensi tool access and balance.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 5 days ago
- Software
- Claude Codeorany agent that can write files
Permissions
File Scopes
Write Files: saves each generated image and its JSON log into ./images/YYYY-MM-DD/ in your project. Terminal: only to create that folder and to save the returned image (decode base64 or download the file link). No other files are read or changed.