Which Sume image models can't edit? Five text-to-image-only rows
Five Sume image rows take text prompts only and reject image_urls: Soul, Imagen 4 Fast and Ultra, Recraft V4 and Qwen Image Max. Prices and what to use instead.

Five Sume image models cannot edit or take references: Soul (higgsfield-soul), Imagen 4 Fast (imagen-4-fast), Imagen 4 Ultra (imagen-4-ultra), Recraft V4 (recraft-v4) and Qwen Image Max (qwen-image-max). In the catalog their image_edit flag is false, and the four with a constraint note say image_urls are not accepted. The other fourteen image rows accept edits or references.
The data is from the image router catalog in Sume's repo, read 2026-10-03. Sume's Image API docs add that a request setting a parameter the model does not list is rejected with 400 unsupported_parameter, so these rows fail fast on a reference.
The five rows
Soul's price is the authenticated Higgsfield estimate for the Sume account, not a published tariff, according to the catalog note. All list prices are before the catalog's times 1.25 rule.
| Id | List price per image | Output formats | Notes from the catalog |
|---|---|---|---|
| higgsfield-soul | $0.004 | provider-selected (PNG) | 1 or 4 images, 720p or 1080p |
| imagen-4-fast | $0.02 | png, jpeg, webp | 5 ratios |
| imagen-4-ultra | $0.06 | png, jpeg, webp | 1K and 2K tiers |
| recraft-v4 | $0.04 | webp only | 13 ratios incl. 21:9 |
| qwen-image-max | $0.075 | png, jpeg, webp | 13 ratios incl. 21:9 |
What to use when you need an edit
Match the replacement to the reason you picked the text-only row:
- Wanted Qwen quality with references: use
qwen-image($0.02 list), which lists edits. - Wanted a source photo in the loop: no Imagen row edits; Nano Banana 2 and Pro list edit support and a
0.5Kto4Ktier. - Wanted Recraft-style graphics: Recraft V4 is text only here, so build the graphic first and edit it in a row that takes references.
- Wanted fine control of a region: ChatGPT Image 2.5 lists an optional
mask_url.
When text-only is the right call
A reference-free row is a fine choice for concept art, backdrops and first drafts, where there is no source picture to preserve. Qwen Image Max at $0.075 list is the priciest of the five, and Soul at $0.004 is the cheapest by a wide margin, so the five rows span almost a twentyfold price range.
Because none of them accepts references, they also cannot be fed a previous output for a second pass. If your workflow loops (generate, then refine), start on a row that takes image_urls.
What Sume does and does not do
Read GET /v1/images/models for the live flags; rows are added over time.
- Does: list
image_editper model so you can filter before you send. - Does: return
400 unsupported_parameterwhen a parameter is not listed. - Does not: serve a reference-guided result from these five rows.
Sources
Related posts
More in Models
- Which Sume image models make 2K or 4K output, by model
FLUX 3 Image added 4K; Sume's catalog has two ways to ask for big images, a resolution tier or custom pixels. Which models take which, and the 3840 edge cap.
- Which Sume video model fits your inputs: text, photo, clip, audio
Match the input you hold to a Sume video model: prompt, first frame, end frame, references, audio sample, or a clip to edit. With the 400s each mix causes.
- An OpenRouter-compatible video API: sume/auto or a pinned model
Sume's POST /v1/videos follows OpenRouter's video generation API field for field. Let sume/auto pick the model, or pin a catalog id like seedance-2.5.
- Image generation API with reference images: POST /v1/images
Send a prompt plus public HTTPS reference images to Sume's POST /v1/images. Pin a catalog model or send sume/auto; the catalog lists each model's limits.
Written by Sume