One hero image to 3D toy, anime and pixel sprite via API
Recraft's Oct 9 art style challenge as three Sume calls: attach the hero, fix silhouette and palette, vary only the medium. Ratios and costs.
To turn one character image into a 3D toy, an anime key visual and a pixel sprite, send the same hero image as an input_references entry in three separate image calls and change only the medium in the prompt. On Sume that is three calls to google/nano-banana-2.1 at 1K, which costs $0.30 in total. Describing the character in text alone will not hold the identity, so the reference does the work.
The method comes from Recraft's Character in Different Art Styles guide, published October 9, 2026 (read 2026-10-13). It runs the challenge in Recraft Studio's Edit image feature with Nano Banana 2.1. The same model is listed on Sume, so the recipe transfers to the Image API.
What does Recraft say stays fixed and what changes?
The guide splits every prompt into two lists. Fixed elements are the silhouette, color palette, signature items and companion character. Variable elements are line quality, shading, lighting, texture and background. It also states that describing a character in text alone produces a cousin, not the same person, and that the hero is attached as image 1.
Prompts also name the medium with concrete traits such as materials, lighting and line weight, not just the word anime or 3D.
| Style | Ratio in the guide | Ratio on Sume nano-banana-2.1 | Vary in the prompt |
|---|---|---|---|
| 3D collectible toy | 3:4 | 3:4 listed | Material, soft studio lighting |
| Anime key visual | 3:4 | 3:4 listed | Line weight, cel shading |
| Pixel sprite | 1:1 | 1:1 listed | Palette size, flat background |
How do I run the three styles on Sume?
Loop over a small table of prompts and keep the fixed clause identical in each. The ratios 3:4 and 1:1 are in the model's catalog list, and input_references accepts up to 10 images, so you can add a second reference later if the face drifts. The script raises on a 202, meaning the call became a job; see the Image API docs for the job envelope.
import os, requests
URL = "https://api.sume.com/v1/images"
HEAD = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
HERO = "https://example.com/hero.png"
KEEP = ("Keep the exact silhouette, color palette, signature items "
"and companion from the reference image.")
STYLES = {
"toy": ("3:4", "Render as a glossy vinyl collectible toy, soft studio light."),
"anime": ("3:4", "Redraw as an anime key visual, clean ink lines, cel shading."),
"sprite": ("1:1", "Redraw as a 64-pixel game sprite, 16 colors, flat background."),
}
def render(ratio, medium):
body = {
"model": "google/nano-banana-2.1",
"prompt": f"{KEEP} {medium}",
"input_references": [{"type": "image_url", "image_url": {"url": HERO}}],
"aspect_ratio": ratio,
"resolution": "1K",
}
r = requests.post(URL, headers=HEAD, json=body, timeout=60)
r.raise_for_status()
return r.json()["data"][0]["url"]
for name, (ratio, medium) in STYLES.items():
print(name, render(ratio, medium))Will the pixel sprite be real pixel art?
No. An image model draws pixel-looking art on a normal canvas, so the grid is usually not exact. If you need a true low-resolution sprite, shrink and quantize the result afterwards; the quantize-to-16-colors recipe does that with Pillow, and the sprite-sheet prompt guide covers animation frames.
How do I check that the identity held?
Recraft ends its guide with four identity-preservation questions. Use your own short version after each render and rerun only the failing style.
- Does the silhouette match the hero at thumbnail size?
- Are the palette and signature items the same?
- Did the face change in the stylized render? If so, add a second reference such as a close-up crop.
- Did lighting shift the colors? If so, name the base colors again in the prompt.
What does it cost on Sume?
Nano Banana 2.1 at 1K is $0.10 per image on Sume (the $0.08 provider list times 1.25), so three styles are $0.30 and a retry of one style is another $0.10. A failed generation uses no credits. Resolution 2K is $0.15 per image if you want print-size art. For a turnaround or expression sheet from the same hero, see the character turnaround recipe.
Sources
Related posts
More in Use cases
- Pick an AI video model by the input you have: photo, clip, audio, none
Match the Sume video row to your input: nothing gets Wan 3.0 or Omni, a photo gets a first-frame row, a clip gets Omni edit or Recast. Read 2026-10-13.
- Place your character in concept art scenes with two references
Design locations first, then put the hero in each one with two input_references. Recraft's Oct 9 method as a Sume API script with costs.
- Poppy Appeal 2026 collection clip: 22 Oct launch to 8 Nov cues
A local group's Poppy Appeal reminder clip: dates from the Royal British Legion's page, then a photo-to-video job and authored cues on Sume.
- St Nicholas Day 2026 is Sunday Dec 6: a 9:16 shoe-and-treat clip
St Nicholas Day is Sunday 6 Dec 2026, with the eve on Saturday. A work-back calendar and a Sume photo-to-video request for a 6-second 9:16 promo clip.
Written by Sume