grok / v1.5 / referenceToVideo

This is the Grok node that does something the v1.0 trio can't: instead of one start frame, it takes up to seven reference images that drive style and content, and you tag them by name in the prompt. Feed it a character sheet, an outfit reference, a location shot - then write "the figure from <IMAGE_0> walks through the street from <IMAGE_2>" and the model assembles a clip from that. It's the closest thing to a character-consistency workflow this pack offers on the Grok side, and the tagging mechanic is the part people get wrong.

Part of civitai-comfy-nodes, Civitai's official ~160-node pack generated from its Orchestration API spec. Cloud job, like the rest: submit a videoGen tagged grok / v1.5 / referenceToVideo, upload the reference images, long-poll, get a native VIDEO tensor back. Billed in Buzz.

The input that matters: images, and how you reference them

images is required and takes 1–7 references through an IMAGE socket. Per the author's tooltip: "tag them in the prompt as <IMAGE_0>, <IMAGE_1>, etc." That's the whole trick - the references are anonymous unless you name them. Don't tag at all and the model just... absorbs them as vibes; tag them explicitly and you can direct which reference controls what. This is the difference between "a video inspired by these" and "this character doing that thing in that place."

• prompt (required, multiline) - use the <IMAGE_N> tags inline.
• duration - 1 to 15 seconds, default 8 (a hair longer than the other Grok nodes' 6).
• resolution - default 480p here, options up to 1080p. 480p is a deliberate cheap-draft default for iteration.
• aspect_ratio - 16:9 default, or 4:3, 3:2, 1:1, 2:3, 3:4, 9:16.

Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING).

Install and auth

Standard for the pack - official, Comfy Registry:

# ComfyUI Manager: search "Civitai Comfy Nodes"
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Restart, authenticate via a Civitai Auth node, CIVITAI_API_TOKEN, or sidebar OAuth.

Common issues

Not tagging the references is the #1 failure mode - the images load fine, the prompt runs, and the output has no idea which reference was which. Write the tags into the prompt, every time. More references isn't automatically better. With seven images the model has to decide what matters; for character consistency, one clean character sheet beats three cluttered stills. Cost compounds - per-second billing plus a longer default duration plus retries adds up; the 480p default exists for a reason, use it for drafts. Your references leave your machine to a third-party server. And the pack is early preview, so node behavior can change without notice. Worth knowing too: this is the Grok node most likely to be confused with the v1.5 imageToVideo node in the same menu - that one treats images as start frames, this one as taggable references. They're not interchangeable.