LTX 2.3 / firstLastFrameToVideo

The frame-anchored member of the LTX 2.3 family: give it a first frame, a last frame, or both, plus a prompt, and it generates the video that plays between them. It runs on Civitai's cloud through civitai/civitai-comfy-nodes, billed in Buzz, no local model.

This is the node for scenes with a defined arc - an object that starts on the left of frame and ends on the right, a camera that pushes in from a wide shot to a close-up, a character that walks from doorway to doorway. Instead of trusting a prompt to land the ending, you show the ending and the model fills the middle. The KB's LTX material even flags a matching pattern in the community's local workflows: frame conditioning is one of LTX's strengths, and first/last frame guidance is how you get a scripted beat to actually stick.

How it works

Standard orchestration plumbing: videoGen workflow with the LTX 2.3 engine, submit, long-poll, download as VIDEO. The guide frames are passed as conditioning (per the tooltips, at least one of first_frame or last_frame must be provided), and frame_guide_strength controls how hard the output holds to them. Stage-wise, the tooltips describe frame-guide conditioning with a strength between 0 and 1 - higher values glue the video to your frames, lower values let the model improvise between them.

Inputs and outputs that matter

• prompt (required) - the action between the frames.
• first_frame - the opening guide image (string input; accepts URL/DataURL/Base64 per the pack's convention).
• last_frame - the ending guide image.
• frame_guide_strength - 0–1, default 0.7. The main creative dial: crank it to hold your frames hard, lower it for more freedom.
• negative_prompt - what to avoid.
• duration - tooltip: 3 or 5 seconds.
• width / height - default 1280×720; divisible by 32.
• fps - default 24; steps - default 20 (8–50); guidance_scale - default 4; generate_audio - default true.
• model - 22b-dev / 22b-distilled.
• loras - CIVITAI_LORAS socket; diffusion_model - transformer override for fine-tunes.
• quantity - 1–10 videos per job, distinct seeds.
• api_config - auth override.

Outputs: video, workflow_id, raw_json.

Installing it

ComfyUI Manager → Custom Nodes Manager → "Civitai Comfy Nodes" (publisher civitai) → Install → restart, or clone + pip install -r civitai-comfy-nodes/requirements.txt. Only requests.

Common issues

• Auth & Buzz. Civitai account with Buzz; CIVITAI_API_TOKEN for headless.
• Both frames missing = server error. The tooltip is explicit: at least one of first_frame or last_frame must be provided. Neither wired, the job fails on the provider side.
• Weak anchoring. If the middle ignores your frames, raise frame_guide_strength toward 1.0; if it's overfitting and looks like a morph, back it off.
• Frame style mismatch. Frames in wildly different styles confuse the model - keep the two frames visually consistent (same subject, lighting, color grade).
• Early preview. Behavior may change without notice; pin the version.
• Cost. Full cloud video job per run; read the per-run cost on the node.