ByteDance Reference Images to Video

Before the fancy Seedance 2.0 multimodal node existed, there was this: the straightforward "give Seedance one to four reference images and a prompt, get a video" node. It's still here, it still works, and for a lot of jobs it's the better tool - fewer moving parts, a cheaper default model, and the same character-consistency trick that made reference-to-video the hot capability of 2026. If you want a subject to stay your subject across the clip, this node's whole design is about that.

It's an API node like everything else in the built-in partner/video category: ships with ComfyUI core, uploads your images to Comfy Org, calls ByteDance's Seedance 1.0 API, polls, and returns the finished clip as VIDEO. No local model files, billed to your Comfy account in credits.

The inputs that matter

• model - defaults to seedance-1-0-lite-i2v-250428, the cheaper/faster lite reference model. The alternative is seedance-1-0-pro-250528, which is pricier but noticeably better on detail and motion. For tests, keep the lite; for anything you'd actually show someone, weigh the pro.
• images - one to four reference images. This is the consistency engine. Multiple angles of the same character or subject is exactly what this node wants.
• prompt - the text description of the video. Important: do not type settings into it. The node appends --resolution, --ratio, --duration, --seed, and --watermark flags to your prompt before sending, and if it detects any of those flag patterns already in your text it raises an error telling you to use the widgets instead. It's a genuinely helpful guardrail, but it will surprise you the first time.
• resolution - 480p or 720p. aspect_ratio - adaptive plus the usual 16:9, 9:16, 4:3, 1:1, 3:4, 21:9. duration - 3 to 12 seconds.

Optional: seed (re-run toggle, non-deterministic results regardless) and watermark (adds an "AI generated" marker, off by default).

Where people get burned

• The flag-in-prompt error. It reads like a broken node the first time you hit it. It's not - it's Seedance's parameter syntax leaking into a field that now has proper widgets. Use the widgets.
• Image constraints are real. Each reference must be between 300×300 and 6000×6000 with an aspect ratio between 0.4 and 2.5. A panoramic 4:1 crop will fail validation.
• It's the older line. Seedance 1.0 is visibly behind 2.0 on things like video editing and audio. If you need those, the Seedance 2.0 Reference node is the upgrade path; if you just need consistent reference-to-video on a budget, the lite model here is genuinely cheap.

One light aside: the display name says "Reference Images" plural because it can, but most people feed it a single strong image and it handles that fine too. The plural is a ceiling, not a requirement.