Precise image editing on Sume: what it has and what it lacks
A plain list of what Sume offers for precise edits after Flux 3 Image and Ideogram 4.5: no Flux 3, no box field, mask_url on one model, Ideogram 4.5 present.

Sume has Ideogram 4.5 and GPT Image 2.5 for precise edits, mask_url on GPT Image 2.5, and up to 16 reference images on that model. It does not list Flux 3 Image, has no bounding-box field, and documents no guarantee that unchanged pixels stay identical on any model. If your job is "change one thing and leave the rest", you can do it with a mask, a crop-and-stitch step of your own, or a careful prompt, and you should verify the result with a pixel diff.
The matrix
| Feature | Flux 3 Image (vendor) | Ideogram 4.5 (vendor) | On Sume |
|---|---|---|---|
| Model in the Sume catalog | Yes, from BFL | Yes, ideogram/ideogram-v4.5 | Flux 3 Image no; Ideogram 4.5 yes |
| Box placement on a 0-1000 grid | Yes, with element ids | Not stated | No field; use prompt words |
| Unchanged-pixel claim | Yes, published shares | Yes, copies untouched pixels | No documented guarantee |
| Mask | Not stated | Yes, up to 3 references with a mask | mask_url on GPT Image 2.5 only |
| References | Up to 10 | Up to 4 | GPT Image 2.5 up to 16; Ideogram 4.5 first plus up to 4 |
| Largest output | About 16.8 MP native | Native 2K | GPT Image 2.5 up to 8,294,400 pixels |
What you can still do
Start from the job, not the model name. A catalog edit of 300 product photos has different needs from one hero image for a campaign: the first wants a cheap tier and a loop, the second wants the best single result and a person looking at it.
For region control, use mask_url on openai/gpt-image-2.5; the docs do not say which mask colour marks the edit area, so test on a cheap render. Ideogram's own page says its masks are black where the edit goes, but that is the vendor's interface, not the Sume parameter. For pixel safety on any model, edit a crop and stitch it back, as in crop, edit and stitch in Pillow, so everything outside the box is the original file.
For text edits, Ideogram 4.5 is the model whose vendor page is about keeping lettering. For repeated edits, save PNG between passes and measure drift with the Pillow script. A failed generation is not billed under the Sume docs, and a 400 unsupported_parameter is returned if you send a field the model does not list.
What it costs
Ideogram 4.5 is $0.0375 (low), $0.075 (medium) or $0.275 (high) per image on Sume, from the provider list times 1.25. GPT Image 2.5 is token priced: $30 per million output tokens, $8 per million image input tokens, $5 per million text input tokens at list, so at 1024x1024 xhigh output is $0.09366 and max $0.21072 before input tokens and Sume's margin.
The Flux 3 launch carries a 50% API discount through October 8 on BFL's side. It does not apply to Sume, since Sume does not serve the model.
Decision in four lines
The earlier Flux 3 alternatives feature by feature goes through each Flux 3 claim in turn.
- Need box-by-box placement: use Flux 3 Image on BFL's service; Sume cannot.
- Need a region fence on Sume: GPT Image 2.5 with
mask_url. - Need text edits and stacked passes: Ideogram 4.5, tested with your own diff.
- Need certainty about every pixel: crop, edit, stitch, and measure.
Sources
Related posts
More in Comparisons
- Real-time or async avatar: six questions to ask before you choose
Need an avatar that talks live, or a finished clip? Six yes/no questions decide it, with what Sume's async avatar jobs cover and what they do not.
- Recraft V4 or FLUX.2 Pro for an illustration set on Sume: price, edits
Recraft V4 bills $0.05 per image on Sume and is text-only; FLUX.2 Pro bills $0.0375 and takes reference images. Ratios, n limits and which fits a set.
- Runway Dev usage exception request vs Sume 429 and 402
Runway Dev says higher usage needs an exception request. On Sume a burst hits 429 or 402. What each means for a batch of ad variants and how to retry.
- Runway Dev image_to_video, video_to_video: one Sume route?
Runway Dev names image_to_video, video_to_video and text_to_image paths. Sume's POST /v1/videos infers the mode from your fields. What it changes in ad code.
Written by Sume