Superside Seedream V5 Pro Edit
Change One Element, Keep the Rest of the Frame Intact
- image_1
- image_2
- image_3
- image_4
- image_5
- image_6
- image_7
- image_8
- image_9
- image_10
- images
- info
Some edit models treat "change the cup" as permission to quietly redo the whole table. Seedream V5 Pro is ByteDance's answer to that: grounded, region-precise editing where you ask for one element to change and the rest of the frame is supposed to stay put. This node wraps it (bytedance/seedream/v5/pro/edit on fal.ai) with up to ten reference images, so it's the pack's "surgical, edit-this-one-thing" node.
What it is and when to reach for it
If the job is a targeted swap - the red car becomes blue, the label changes, one product in a lineup gets a new finish - V5 Pro is the natural pick, and the README's framing ("changes one element while keeping the rest of the frame intact") is the pitch. Its sibling V4.5 is the broader multi-reference editor with a higher 4K ceiling; V5 Pro trades some of that headroom for precision and a couple of workflow niceties like an explicit output format. In a production pipeline, V5 Pro is the "final retouch" node you slot after a cheaper draft pass.
The inputs that matter
prompt+image_1+api_key- the required trio;image_2…image_10for up to ten references.image_size-auto_1K/auto_2K(default) presets, plus square/portrait/landscape shapes. Note: noauto_4Khere, unlike V4.5 - V5 Pro tops out around 2K.size_mode/width/height- custom mode is available, and the tooltip spells out the constraint: total pixels must stay between 1024×1024 and 2048×2048, aspect ratio between 1/16 and 16. Stay in preset mode unless you have a specific target.output_format-jpeg(default) orpng. Choose png if downstream compositing needs lossless.num_images- 1–6, number of variations to request.enable_safety_checker- on by default.sync_mode- off by default. This node runs through fal's queued execution path (polling) because these multi-minute edits drop long connections; leave it as-is.
Outputs
images(IMAGE) - results back in the graph as normal tensors.info(STRING) - the result URL.
Installing it
It's part of comfyui-superside-nodes - clone, install requirements, restart:
cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes.git
cd comfyui-superside-nodes
pip install -r requirements.txt
Paste your fal api_key. No model files - Seedream runs on fal's servers, every call is metered, and your images + prompt leave your machine.
The honest take
"Grounded, region-precise" is a promise, not a guarantee - seedream-grade edits hold the frame much better than older models, but "change the cup" can still drift the tablecloth when the reference is low-res or the element blends into its surroundings. Feed it the highest-quality reference you have and keep the prompt to one clear change; that's where it shines. And if you find yourself re-rendering because a neighbor element shifted anyway, that's the point to reach for a masked local inpaint instead - the model you can run locally is the cheaper fix when the region is small and precise. V5 Pro earns its call when the whole frame has to survive.
Inputs (20)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| image_1 | IMAGE | — | |
| api_key | STRING | — | |
| image_2opt | IMAGE | — | |
| image_3opt | IMAGE | — | |
| image_4opt | IMAGE | — | |
| image_5opt | IMAGE | — | |
| image_6opt | IMAGE | — | |
| image_7opt | IMAGE | — | |
| image_8opt | IMAGE | — | |
| image_9opt | IMAGE | — | |
| image_10opt | IMAGE | — | |
| size_modeopt | COMBO | preset | 2 options: preset, custom |
| image_sizeopt | COMBO | auto_2K | 8 options: auto_1K, auto_2K, square_hd, square, portrait_4_3, portrait_16_9, +2 |
| widthopt | INT | 102464–2048 | Used when size_mode is custom. Total pixels (width x height) must stay between 1024x1024 and 2048x2048, with aspect ratio between 1/16 and 16. |
| heightopt | INT | 102464–2048 | Used when size_mode is custom. Total pixels (width x height) must stay between 1024x1024 and 2048x2048, with aspect ratio between 1/16 and 16. |
| output_formatopt | COMBO | jpeg | 2 options: jpeg, png |
| num_imagesopt | INT | 11–6 | — |
| enable_safety_checkeropt | BOOLEAN | true | — |
| sync_modeopt | BOOLEAN | false | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| info | STRING | — |