flux / v3.0 / firstLastFrameToVideo

Here's the ComfyUI-native version of a trick you've seen everywhere else: give a model the first frame and the last frame, and let it dream up everything in between. It's how you animate a still, morph between two poses, or turn a before/after pair into a transition - and it works especially well when the two frames come from the same image generator, because the model isn't fighting a style change. This node runs FLUX 3's first-and-last-frame video on Civitai's fleet, with native audio.

What it is

flux / v3.0 / firstLastFrameToVideo from civitai/civitai-comfy-nodes, under Civitai/Video/flux. Two IMAGE sockets in, a VIDEO out, a closed-model cloud job in between. FLUX 3 is BFL's multimodal model (image/video/audio in one), and it's API-only - no open weights - so hosted is the only way to run it, and Civitai is the intermediary that drops it into your graph.

The inputs

• prompt - required. Describes the motion between the frames. "She turns to the camera," "the car drifts through the corner," "the flower blooms" - the prompt is your director's note for the in-between.
• start_image / end_image - both required IMAGE sockets. The tooltip notes they can be "A URL, A DataURL or a Base64 string" - the pack converts whatever IMAGE you wire in, so a normal Load Image output just works. Keep the two frames visually consistent: same subject, same style, different state. Wildly different frames make the interpolation the model's problem, and the model's problem becomes your problem.
• duration - 5–20 seconds, default 5.
• resolution - 720p or 1080p, default 720p.
• aspect_ratio - auto plus the standard set (16:9, 1:1, 9:16, etc.).
• generate_audio - default true. FLUX 3 does native synchronized audio; turn it off for a silent clip.
• draft - default false. Draft mode renders a cheap 720p version carrying a draft cache; feed that to the flux / v3.0 / draftEnhance node for the full-quality pass. For anything you're iterating on, draft first.

The outputs

• video - the interpolated clip, as a VIDEO.
• draft_cache_audio / draft_cache_video - the caches when draft is on, ready for draftEnhance.
• workflow_id / raw_json - the pack standard.

Installing it

ComfyUI Manager → Civitai Comfy Nodes → Install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Gotchas

• The frames decide everything. Two consistent frames give you a tight interpolation; two unrelated ones give you a morph that looks like a different video. Make the start and end frames deliberately, don't grab random outputs.
• Metered per second. 1080p at 20 seconds with audio is a real charge. Draft mode is the escape hatch - and with a first/last-frame workflow, the draft is genuinely worth reviewing before you spend the full pass.
• Your images and prompt leave the machine. Closed model, hosted call, by design.
• Credentials - Civitai Auth node, CIVITAI_API_TOKEN, or sidebar OAuth. No token, no interpolation.
• The 5-second floor on duration is the model's minimum, not a suggestion - you can't request a 2-second blink-and-miss clip.