Tell it what to change — clothing, pose, light, mood — one instruction at a time.
Spicy 1.0 Image Edit (Qwen Image Edit Spicy) applies precise, instruction-based edits to any image: remove or change clothing, adjust pose and expression, relight the scene, swap backgrounds — while explicitly preserving everything else. Supports (keyword:weight) emphasis syntax for fine control. The refinement stage of every Spicy pipeline.
No output yet. Run the model to see results.
Click 'Run Model' to generate output
Exact unit prices and billing rules for this model. All prices in USD.
Endpoint
https://api.spicyrouter.ai/v1/imagesAll requests require a Bearer API key.
Submit Task
curl -X POST \
"https://api.spicyrouter.ai/v1/images" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "qwen-image-edit-spicy",
"image_url": "https://example.com/your-asset",
"prompt": "your prompt here"
}'Poll Status
# Replace TASK_ID with the data.task_id from the submit response
curl -X GET "https://api.spicyrouter.ai/v1/status/TASK_ID" \
-H "Authorization: Bearer YOUR_API_KEY"Request Parameters
| Field | Required | Type | Values | Description |
|---|---|---|---|---|
image_url | Yes | string | Default: | Image Url |
prompt | Yes | string | 1 ~ 4000 chars Default: | Prompt |
Battle-tested prompts to get production-grade output in one shot.
Enhance photorealism. Add natural skin texture, pores, and subtle imperfections. Keep pose unchanged.
Change lighting to soft golden-hour from the left side. Maintain current pose and expression.
Replace background with a luxury hotel bedroom. Keep the person exactly as is.
Why Spicy Image Edit
Regenerating a whole image to fix one detail is a tax on your budget and your patience. Spicy Image Edit changes only what you tell it to — and keeps everything else pixel-faithful.
Write a modification, not a scene description. Change her dress color from blue to red. Keep everything else unchanged. — the model edits, it doesn't reimagine.
Dial visual emphasis up or down with (keyword:1.5) to strengthen and (keyword:0.7) to soften — fine-grained control over what dominates the frame.
Explicitly lock what must not move — face, expression, pose, background. Identity stays intact across every iteration, which is exactly what series content needs.
Editing Principles
An edit prompt is a modification order, not a full scene description. ✗ A beautiful woman with long hair wearing a red dress in a bedroom → ✓ Change her dress color from blue to red. Keep everything else unchanged.
Iterate in layers: pass 1 adjusts clothing → pass 2 enhances realism → pass 3 relights. Several small edits beat one ambitious overhaul every time.
Spell out the preserved regions: Make her clothing more revealing. Keep her face, expression, pose, and background exactly the same.
Enhance photorealism. Add natural skin texture, pores, and subtle imperfections. Keep pose unchanged. / Change lighting to soft golden-hour from the left side. Maintain current pose and expression. / Replace background with a luxury hotel bedroom. Keep the person exactly as is.
Command Reference
Battle-tested instructions for the three most common edit categories. English input gives the best results.
remove all clothingchange outfit to black lace lingerieadd a white silk robe loosely drapedadjust pose to lying on backchange expression to seductive smileadd sweat droplets on skinenhance muscle definitionchange background to luxury hotel roomadd warm candlelight atmosphereadd steam/fog effectWeight Syntax
Wrap any keyword in parentheses with a numeric weight. Above 1.0 pushes it forward, below 1.0 pulls it back, no parentheses means neutral 1.0.
1girl, solo, nude, kneeling on bed, (ass focus:1.5), (wet skin:0.7), soft warm lighting, looking over shoulder
(keyword:1.5) strengthens — the higher the number, the more prominent the element. (keyword:0.7) weakens — the lower the number, the more it fades. No parentheses is neutral 1.0; the example pushes the rear focus hard while dialing the wet-skin sheen down.
FAQ
You can, but you shouldn't. One direction per pass — clothing first, then realism, then lighting. Chained single-purpose edits preserve far more of the original than one overloaded instruction.
State it explicitly in every instruction: Keep her face, expression, pose, and background exactly the same. Naming the preserved regions is the single most reliable identity anchor.
English. Instruction following is measurably more precise with English prompts, especially for preserve-region clauses and weight syntax.
Whenever the frame is 80% right. Regeneration re-rolls everything; editing fixes only the 20% that's wrong — and keeps the composition you already approved.
Take the frame you love and make it exact — one instruction, one iteration, one flawless image at a time.