Seedream 5.0 Lite · Edit

Lite's pitch is boring and useful in equal measure: it edits from source images at a fixed 2K output size, chosen from a short list of exact resolutions, every time. No auto-sizing, no tier negotiation, no "it came back 1.4 megapixels when I asked for 2". If you're producing a series - catalogue shots, a set of thumbnails, storyboard frames - the thing that ruins a batch is drift in dimensions, and this node removes that variable.

5.18 credits (~$0.052) per image. More than Flash, less than Pro.

How it works

PopNow's node for bytedance/seedream-5.0-lite/edit. image_urls is an autogrow socket group - connect a Load Image, get another socket, as many as the model accepts. Your sources upload on queue (content-hash cached for 12 hours), the request goes to ByteDance's model on PopNow's servers, and the results come back as a ComfyUI IMAGE batch.

The tooltip on prompt is the tell for how to prompt it: describe the image to produce from the source - subject, composition, style, lighting. Not the change. Write the whole target picture and let the model work backwards from the reference.

The inputs

• image_urls - the source images. Multiple allowed, so character plus prop plus background all at once.
• prompt - the target image, in full.
• size - width*height from a fixed list: 2048*2048 (default), 2304*1728 / 1728*2304, 2848*1600 / 1600*2848, 2496*1664 / 1664*2496, 3136*1344. Eight options, all in the 2K range, square through ultrawide. Pick one and stay on it for the whole series and nothing downstream will need resizing.
• output_format - jpeg or png. PNG if the frames are going anywhere else afterwards.

Outputs

images (batch) and task id.

Notice what isn't there: no alpha, no background, no seed, no aspect-ratio picker separate from size. Lite's schema is narrower than Flash's and Pro's - if you want transparent output you're on the wrong node, and that's worth knowing before you build a workflow around it.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Or search PopNow in ComfyUI Manager and Install. Then add the key - popnow.ai/api/console/keys into Settings > PopNow, or POPNOW_API_KEY in the environment - and restart a second time so the catalog rebuilds the node list. There's no pip step at all; the pack's pyproject.toml declares no dependencies and relies on ComfyUI's own aiohttp/torch/Pillow/NumPy/PyAV.

Common issues

• size rejected - it's a dropdown with eight entries. 4096*2304 is the 4.0/Pro world, not Lite's.
• The edit ignored half the references - very common with multi-image input. Prompt in slots: "the person from image 1, the jacket from image 2". Vague multi-reference prompts get averaged into mush.
• PopNow 40201 - out of credits. Five credits a frame adds up over a series; do the multiplication first.
• PopNow 40004 - unsupported combination; the message names the field.
• PopNow 40310 - unverified email. Blocks runs, easy to miss.
• Cancelling stops your wait, not the job - it completes server-side and bills on success.

When Lite is the right pick

Between Flash (cheaper, more flexible, has transparency) and Pro (sharper, more references, has a enable_thinking pass), Lite's case is specifically the series job: one fixed size, a reference image per frame, consistency across a set. That's closer to how you'd build a catalogue than to how you'd chase a single hero image, and it's exactly the shape of work this family's middle tier was made for.

If you need a character held across many frames rather than a handful, though, a hosted per-image edit is the wrong tool and a trained character LoRA is still the right one - hosted models don't accumulate a character, they re-derive it from whatever references you paste in each time.