Wan 2.7 Image Spicy
Wan 2.7 Image Spicy keeps the full Wan 2.7 image contract - optional reasoning before rendering, color palettes, bounding-box edits, up to 9 input images, and 4K text-to-image on Pro - on a route with upstream input and output content inspection disabled, for adult-oriented creative work.
Available Models
Wan 2.7 Image Spicy
Wan 2.7 Image Spicy on APIPod is Wan 2.7 Image with upstream content inspection relaxed: text-to-image up to 2K with an optional reasoning stage (thinking_mode) that plans composition before rendering, color palette control, and 1-4 images per request.
Wan 2.7 Image Spicy Edit
Wan 2.7 Image Spicy Edit on APIPod is Wan 2.7 Image Edit with upstream content inspection relaxed: edit or recompose 1-9 input images by instruction, with optional bounding boxes to localize changes and color palette control.
Wan 2.7 Image Pro Spicy
Wan 2.7 Image Pro Spicy on APIPod is Wan 2.7 Image Pro with upstream content inspection relaxed: high-resolution text-to-image with a 4K tier, optional reasoning before rendering, color palette control, and 1-4 images per request.
Wan 2.7 Image Pro Spicy Edit
Wan 2.7 Image Pro Spicy Edit on APIPod is Wan 2.7 Image Pro Edit with upstream content inspection relaxed: higher-fidelity editing of 1-9 input images by instruction, with bounding-box localization and color palette control.
18+Wan 2.7 Image Spicy · 4 models
Your colors, exactly.Filter off.
It's Alibaba's Wan 2.7 Image, with the upstream content check switched off. Give it exact HEX colors and how much of each to use, describe the face shape you want, and box the part of a photo to change. Same model, same API, fewer rejections.
- Models
- 4
- Standard and Pro, each with generate and edit.
- Max output
- 4K
- On Pro text-to-image. Editing goes up to 2K.
- Reference images
- 9
- For multi-subject consistency and compositing.
- Text in image
- 12 langs
- Prompts up to 3K tokens. Readable text, tables and formulas.
- Thinking mode
- On
- Plans layout and lighting before drawing. Text-to-image only.
- Upstream check
- Off
- Input and output inspection disabled on every request.
01 · Palette lab
Pick the exact colors. Set how much of each.
color_palette takes 3 to 10 HEX colors and a share for each. The shares must add up to 100.00%. Choose a preset to see its color mix and the request JSON you'd send. Matisse and Picasso are Alibaba's own examples; the other three are our mature-theme mixes.
Presets
Mostly black and oxblood shadow, rose skin tones in the midrange, a small gold accent from a candle or jewelry. Built for low-key bedroom portraits.
This is our own mix, not an official example. Copy the JSON and pair it with a prompt from the recipes below.
"color_palette": [
{ "hex": "#1A0F12", "ratio": "30.00%" },
{ "hex": "#5C1A2B", "ratio": "22.00%" },
{ "hex": "#B23A48", "ratio": "16.00%" },
{ "hex": "#E8B4A0", "ratio": "14.00%" },
{ "hex": "#F5E6DA", "ratio": "10.00%" },
{ "hex": "#C9A227", "ratio": "8.00%" }
]02 · Faces
Not the same AI face every time.
Wan 2.7 changes the face shape or eyes when you ask, and keeps the person, outfit and scene the same. Choose a trait below and only that word in the prompt changes.
Face shape · oval face
Official Wan 2.7 Image samples. Each tile changes one trait in the prompt.
“Portrait of a young man in a woven tribal jacket in front of a mountain village, oval face, natural light, 50mm”
03 · Box editing
Box what to change. Leave the rest alone.
In the -edit models, bbox_list sets the areas to edit, up to 2 boxes per input image. Whatever is inside the boxes gets replaced, and the face, pose and light outside them stay as they were.
box 1 · outfitbox 2 · backdropEdit instruction
In box 1, change the dress to a black lace slip with thin straps. In box 2, change the lake to a dim hotel room lit by red neon. Keep her face, hair, pose and lighting exactly as they are.
"model": "wan2.7-image-pro-spicy-edit",
"image_urls": ["https://…/selfie.jpg"],
"bbox_list": [[[102, 635, 963, 1014], [10, 20, 307, 553]]]04 · The four models
Standard or Pro, generate or edit.
All four use the same request format as Wan 2.7 Image and run on the same upstream model. The only change is that upstream content inspection is off.
wan2.7-image-spicy
Text to image
- Input
- Text
- Output
- 1K / 2K
- Price
- $0.03 / image
wan2.7-image-spicy-edit
Edit & composite
- Input
- Text + 1–9 images
- Output
- 1K / 2K
- Price
- $0.03 / image
wan2.7-image-pro-spicy
Text to image, up to 4K
- Input
- Text
- Output
- 1K / 2K / 4K
- Price
- $0.075 / image
wan2.7-image-pro-spicy-edit
Pro editing
- Input
- Text + 1–9 images
- Output
- 1K / 2K
- Price
- $0.075 / image
You pay per image returned, so n = 4 costs four times as much. Prices come from the live catalog.
05 · Spicy recipes
Prompts with palettes. Adults only.
Each recipe pairs a prompt with one of the palette presets. Copying gives you the full request body, ready to send.
Candlelit satin
Candlelit boudoir portrait of a woman in a black satin robe on the edge of an unmade bed, one strap slipping off her shoulder, warm rim light on bare skin, 85mm, shallow focus, film grain
Room 9, 2 a.m.
Rain-streaked motel window at 2 a.m., a woman in a lace bralette and an oversized leather jacket leaning against the door frame, pink and cyan neon spilling across her skin, cigarette haze, cinematic still
Velvet curtain call
Burlesque dancer mid-pose on a velvet stage, feather fans covering her chest, rhinestone corset, magenta spotlight, a sliver of crimson curtain, smoky 1950s cabaret atmosphere, medium format
Red room pin-up
Vintage pin-up illustration in Matisse's flat red interior style, a woman in a polka-dot swimsuit reclining on a patterned chaise, teasing over-the-shoulder glance, bold gouache shapes, poster composition
06 · API
One request. Palette included.
Send it to /v1/images/generations and poll the task ID you get back. thinking_mode is on by default for text-to-image.
Read the docscurl https://api.apipod.ai/v1/images/generations \
-H "Authorization: Bearer $APIPOD_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan2.7-image-pro-spicy",
"prompt": "Candlelit boudoir portrait, black satin robe slipping off one shoulder, warm rim light, 85mm, film grain",
"size": "2K",
"aspect_ratio": "3:4",
"n": 2,
"thinking_mode": true,
"color_palette": [
{ "hex": "#1A0F12", "ratio": "30.00%" },
{ "hex": "#5C1A2B", "ratio": "22.00%" },
{ "hex": "#B23A48", "ratio": "16.00%" },
{ "hex": "#E8B4A0", "ratio": "14.00%" },
{ "hex": "#F5E6DA", "ratio": "10.00%" },
{ "hex": "#C9A227", "ratio": "8.00%" }
]
}'
# → { "code": 200, "data": { "task_id": "…" } }
curl https://api.apipod.ai/v1/images/status/$TASK_ID \
-H "Authorization: Bearer $APIPOD_API_KEY"Asking for 4K on wan2.7-image-pro-spicy only works when the request has no input images. aspect_ratio is turned into exact pixel sizes on our side.
Read this before you send anything
18+ only
This series is for adults making mature-theme creative work. Don't integrate it into anything minors can use.
Not in mainland China, Hong Kong or Macau
Requests from those regions are rejected, and the series is hidden from catalog pages there.
Some content is always banned
Anything involving minors, non-consensual sexual content, real people's likeness used without consent, and anything illegal where you or your users are. Upstream inspection being off doesn't change that.
You're responsible for what you publish
Outputs aren't filtered upstream. If you show them to users, you need your own review and age checks.
FAQ
Questions
What is Wan 2.7 Image Spicy?
It's Alibaba's Wan 2.7 Image and Image Pro with DashScope's upstream content inspection turned off for both inputs and outputs. The model, request format and output quality are the same as the standard series.
How is it different from wan2.7-image?
Only in the content inspection. The parameters (size, aspect_ratio, n, thinking_mode, color_palette, bbox_list, seed, watermark) and their limits are the same. Prompts that the standard series rejects for mature themes go through here.
How does color_palette work?
Give it 3 to 10 colors. Each one is a #RRGGBB hex value plus a share written as NN.NN%, and the shares must add up to exactly 100.00%. The model follows how much of each color to use, not just which colors appear. It works for text-to-image and editing, but not group-image mode.
Can I control a face precisely?
Yes. Put the face shape (oval, square, round, long, diamond) and eye shape (almond, deep-set, smiling, puffy, phoenix) in the prompt. Wan 2.7 is trained to stay away from the generic AI face, so specific facial words make a visible difference.
What do the -edit models accept?
From 1 to 9 input images as URLs or data URIs, plus an instruction. You can add bbox_list with up to 2 boxes per image, written as [x1, y1, x2, y2] in pixels, to keep edits inside those areas.
What resolutions are supported?
1K and 2K on every model. 4K is only on wan2.7-image-pro-spicy for text-to-image without input images. You can also set an exact size as width*height, and aspect ratios from 1:8 to 8:1 work.
How is it billed?
Per image returned. With n = 2 you pay for two images. Prices for each model are on its model page. Failed tasks are refunded automatically.
Why can't I use it in my region?
The Spicy series isn't offered in mainland China, Hong Kong or Macau. Requests from there get a 403, and the model pages are hidden.
Want an uncensored image to match your brand colors?
Create a key, choose one of the four models, and add a color_palette to your first request.