Seedream 5.0 Pro on fal: A Cheap Image Editor With Three Tiers, and the Face Question (October 2026)
ByteDance's Seedream 5.0 edits images from up to ten references for $0.027 to $0.0675 per image across Flash, Lite and Pro. What each tier costs, what it accepts, and why we have not put a real face through it.
Seedream is ByteDance's image model family, and the 5.0 generation arrived on fal in July 2026 (sources disagree on the exact date, so we say only "July") with an editing endpoint in three tiers. It is a cheap, capable editor for products, scenes and design work. It is also from the same company as Seedance, whose video models refuse real human faces, which is why a section of this guide is about what we do and do not know about faces.
The three tiers
| Tier | Price per image | Use it for |
|---|---|---|
| Seedream 5.0 Flash | $0.027 | rough compositions, batches of variants |
| Seedream 5.0 Lite | $0.035 | everyday edits and mock-ups |
| Seedream 5.0 Pro | $0.0675 | final-quality results, finer detail |
Read from fal on 18 October 2026. A pack of fifty product variants on Flash is $1.35; the same fifty on Pro is $3.38. Compare that with Nano Banana 2.1 at about $0.034 for a 1K image through Google's API (guide) and GPT Image 2.5 at roughly $0.05 to $0.25 depending on quality (comparison): Seedream's Flash and Lite tiers are among the cheapest ways to run many edits.
What it accepts
- References: up to ten images in one request, so you can supply a product, a setting and a style reference together.
- Sizes: from 1024 by 1024 up to 2048 by 2048 pixels.
- Layerize: an option that splits an image into separable layers, handy for design and ad work where you want to move the headline or swap the background.
The edit endpoint follows a prompt like "replace the background with a sunlit kitchen, keep the bottle exactly as it is", and its strength is staying close to the reference when the instructions are specific.
Where it fits in a video workflow
Images are the cheap half of any video pipeline. The principle we repeat across these guides: decide composition in images, decide motion in video. An approved still costs cents to produce and saves dollars in video retries. Seedream suits the unglamorous jobs inside that principle:
- Product shots. Drop a bottle onto five backgrounds and pick the one worth animating.
- Thumbnails and covers. Flash is cheap enough to try twenty variants.
- Scene plates. An empty set to animate with a face-less image-to-video model.
- Style references for a character series, before you commit to a design.
The face question
This is the part we have not tested and therefore will not claim. ByteDance's video models (Seedance 2.0 and 2.5) reject real human faces, a policy confirmed by the vendor for 2.5 on 31 August 2026 and one we also found on the image endpoints of the Seedance family in our own April testing. Whether Seedream 5.0's edit endpoints accept a real person's photo is something fal's documentation does not settle and we have not paid to find out. If your job needs a real face in an image, use a model with a known record: Nano Banana 2.1 or FLUX.3 edit, and check that you have the rights to the person's likeness.
For the sake of precision: we are not saying Seedream refuses faces. We are saying we do not know, and the vendor's other products do.
When not to use it
Not for the reference images that anchor a character's identity in a video pipeline, because those are exactly the faces whose handling is unverified. Not when you need transparent backgrounds natively (GPT Image 2.5 lists that option). And not as a substitute for an image editor with a layer stack when you have a designer; layerize is a convenience, not a replacement.
Dates and honesty
fal's listing and third-party coverage agree on July 2026 and disagree on the day, and the prices above are the three per-image figures fal showed on 18 October. We read the schema and prices only; we have not generated with this model for this article, so the guide is a specification, not a review.
Common questions
Which tier should I use?
Flash for rough variants, Lite for everyday edits, Pro for final quality. At $0.027, $0.035 and $0.0675 per image the difference is small enough to test all three on one prompt.
Does Seedream accept real faces?
We do not know. ByteDance's video models refuse real faces and we found the same on the Seedance image endpoints, but we have not tested Seedream 5.0 and will not claim either way.
How many references can it take?
Up to ten images in one edit, at output sizes from 1024 to 2048 pixels square.
What does layerize do?
It splits an image into separable layers, which is useful for design and ad work where you want to move text or swap a background.
How to do this without code: Your avatar is built from one photo in the product, and creating it is free. In AutoHustle this is the Me in scene tool: a few photos of you and one sentence describing the scene (about 2 credits per second, quoted before you press Generate). See How to put yourself in any scene and Put yourself in a movie scene.
Related reading: AI Video Models on fal in October 2026: What Is New, What It Costs, and What Is Gone · FlashTalk on fal: A $0.02-per-Second Talking Head That Cannot Make a Vertical Video (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| Seedream 5.0 Pro edit | bytedance/seedream/v5/pro/edit |
| Seedream 5.0 Lite edit | bytedance/seedream/v5/lite/edit |
| Seedream 5.0 Flash edit | bytedance/seedream/v5/flash/edit |
Prices read on 18 October 2026: Pro $0.0675, Lite $0.035, Flash $0.027 per image. Up to 10 reference images, output 1024²–2048², optional layerize.