Every Sume image model compared: edit, text-only, n and formats
Sume's repo catalog lists 20 image rows: 14 take edit references, 5 are text-only, 1 makes layers. Grok is capped at n 1; Recraft is webp only.
The catalog in the Sume repo has 20 image rows as of 2026-10-13. Of those, higgsfield/soul, google/imagen-4-fast, google/imagen-4-ultra, recraft/recraft-v4 and qwen/qwen-image-max are text-to-image only, qwen/qwen-image-layered is layers-from-one-image, and the other 14 take edit references. x-ai/grok-image is capped at n: 1, and every other row allows up to 4.
The full table
Billed per image is list x 1.25 before the per-job round-up to the cent, from the repo's pricing table; sizes and qualities above the default can cost more, so use the rate card for a budget.
| Sume model id | Billed | Edit refs | n | Notes |
|---|---|---|---|---|
| higgsfield/soul | $0.005 | No | 1 or 4 | 720p or 1080p |
| openai/gpt-image-2.5 (+ sunburst) | about $0.066 at high 1024 | 16 | 1-4 | mask_url, background |
| openai/gpt-image-2 | $0.26375 at high 1024 | 10 | 1-4 | low, medium, high |
| google/nano-banana-2.1 | $0.10 | 10 | 1-4 | 0.5K to 4K |
| google/nano-banana-pro | $0.1875 | 10 | 1-4 | 0.5K to 4K |
| bytedance-seed/seedream-5-lite | $0.04375 | 10 | 1-4 | |
| bytedance-seed/seedream-4.5 | $0.05 | 10 | 1-4 | |
| bytedance-seed/seedream-4 | $0.0325 | 10 | 1-4 | lists auto |
| x-ai/grok-image | $0.025 | 10 | 1 | 13 ratios |
| qwen/qwen-image | $0.025 | 10 | 1-4 | |
| qwen/qwen-image-max | $0.09375 | No | 1-4 | |
| qwen/qwen-image-layered | $0.0625 | 1 | 2-4 layers | png |
| google/imagen-4-fast | $0.025 | No | 1-4 | |
| google/imagen-4-ultra | $0.075 | No | 1-4 | 1K, 2K |
| black-forest-labs/flux.2-pro | $0.0375 | 10 | 1-4 | |
| black-forest-labs/flux.2-flex | $0.0625 | 10 | 1-4 | |
| ideogram/ideogram-v3 | $0.075 | 10 | 1-4 | quality maps to speed |
| ideogram/ideogram-v4.5 | $0.075 at medium | 5 | 1-4 | low, medium, high |
| recraft/recraft-v4 | $0.05 | No | 1-4 | webp only |
Reading it
The table shows 19 lines for 20 rows because the two GPT Image 2.5 variants share one line. Soul's rate is an estimate, not a published tariff, per the catalog note.
Always confirm with the live list before you pin an id:
curl -s https://api.sume.com/v1/images/models \
-H "Authorization: Bearer $SUME_API_KEY"Quick picks
- Cheapest edit-capable: Qwen Image or Grok Image at $0.025.
- Most references: ChatGPT Image 2.5 at 16.
- Widest ratio lists: Nano Banana 2.1 and Ideogram, 15 values each.
Sources
Related posts
More in Models
- H3 Max lip sync and audio over 14.8 seconds: split or Fabric
Audio over 14.8 seconds is refused on Sume's H3 Max lip sync because the provider silently clips the rest. Fabric 1.0 takes the long line.
- H3 Max lip sync rejects audio under 5 seconds: what to do
Sume's MiniMax H3 Max lip sync takes 5 to 14.8 seconds of audio and returns invalid_request outside that window. Fabric is the fallback for shorter lines.
- Porting a Sume agent loop to Haiku 5.5: drop temperature, top_p, top_k
Anthropic says to omit temperature, top_p and top_k on Haiku 5.5, and some values return a 400. Strip them from your loop before the first Sume tool call.
- HelixBench: what spatial-acoustic consistency tests in AI video
HelixBench scores whether synthesized sound fields track camera movement. What the arXiv page says it measures, and a manual way to spot-check any AI clip.
Written by Sume