Google AI - Video Storyboard

Storyboard is the right name for this node: instead of generating a clip from a bare prompt, it takes up to three reference images and uses them to steer the video. That's the difference between "a stylized animated scene" and a clip that actually holds onto your character, your location, or your art style from still to still. GoogleAI_VideoStoryboard is the Veo 3.1 node in this pack for anyone who cares about consistency rather than just vibes.

What it's for

Keeping a subject recognizable across the generated motion. You've got a character design, a product shot, or a location you've already made as an image - drop one or two references in, and the video should respect them instead of drifting into generic AI footage. It's the closest this pack gets to character consistency in cloud video, and it's a genuinely different workflow from plain text-to-video: plan the stills first (locally, cheaply), then animate.

How it works

The reference images are sent to Veo alongside your prompt, and the model generates a clip that starts from that visual context. One implementation detail worth knowing: when references are present, the duration is forced to 8 seconds - the tooltip says so explicitly - so your duration_seconds choice only really applies when you're generating without references. The clip returns as a video_frames tensor with native audio on Veo 3.1, plus a cost_estimate string, and errors degrade to an error image rather than a crash, per this pack's design.

Inputs and outputs that matter

• prompt (STRING, multiline, default "A stylized animated scene with vibrant colors") - the action and mood on top of the references.
• reference_image_1 … reference_image_3 (IMAGE) - optional but the whole point. One is fine; three is a stronger storyboard.
• model (COMBO, default veo-3.1-generate-preview) - Veo 3.1 or the fast variant.
• video_preset (COMBO) - resolution/aspect presets, default 1080p 16:9.
• duration_seconds (COMBO) - effectively ignored when references are attached (forced to 8s); set it for reference-free runs.
• video_frames (IMAGE), audio (AUDIO), cost_estimate (STRING) - the outputs.

Installing it

One of the 15 Google nodes in COMFYUI_PROMPTMODELS (PromptModels Studio in Manager):

cd ComfyUI/custom_nodes
git clone https://github.com/cdanielp/COMFYUI_PROMPTMODELS

Add GEMINI_API_KEY=AI... to the pack's .env and restart. Needs ffmpeg on the system and ComfyUI 0.26.0+.

Common issues

The forced-8-seconds-with-references behavior surprises people - you pick 4 seconds, attach a reference, and get an 8-second clip with no warning beyond the tooltip. That's by design, not a bug. Reference steering is also proportional to how distinct your images are: a vague reference pulls the video less than a clear, well-lit one. And remember the meter is running: storyboard runs are billed Veo calls, and the cost_estimate output exists so you can watch the spend before you queue a batch of ten.