Convert a non-24 fps reference video to 24 fps before loading it into H3 R2V. The ComfyUI documentation says R2V reads reference video at 24 fps and does not use the source frame-rate metadata. It also treats picture and soundtrack as separate references, so an accompanying audio track must be connected to the matching ref_video_N through ref_video_audio_N. This preparation is for the model’s reference input, not a decision about the frame rate of your final editing timeline.
When conversion is needed
Use this preparation when the reference file’s actual frame rate is not 24 fps, including common rates such as 30 or 60 fps. Do not assume the file will be interpreted at its recorded rate just because a media player reports that rate or the file contains frame-rate metadata. The documented R2V behavior is to read reference video at 24 fps without that metadata.
A frame-rate conversion changes how frames are sampled or timed to make a 24 fps output; the exact conversion method and its visual consequences depend on the tool and settings you choose. The ComfyUI page specifies the required rate but does not prescribe a converter, conversion algorithm, or an audio synchronization check. Keep those as separate preparation and review tasks.
Prepare and map the references
- Read the source file’s frame rate in your media tool and record the value as displayed. Distinguish exact 24 fps from nearby rates such as 23.976 if your tool reports that precision.
- Convert the reference video to an output reported as 24 fps. Keep the original as your source; use the converted file as the R2V video reference.
- In the ComfyUI R2V workflow, place videos in the intended order and note each
ref_video_Nindex. Reference inputs follow the documented category order: images, then videos, then standalone audio. - If a video has a soundtrack you want to use, connect it separately to the corresponding
ref_video_audio_N. Do not treat the video index and audio index as one shared numbering system: the documentation says picture and audio indices are independent, and video soundtracks are numbered first in video order before standalone audio clips. - Check the graph wiring and reference labels before generation. If there are standalone audio clips as well, record their separate audio indices so a video soundtrack is not mistaken for one of them.
The documented R2V limits are up to three videos and three standalone audio clips, alongside up to nine images. The guide also notes that video soundtrack inputs need both vae and audio_vae; standalone audio needs audio_vae only. See the official H3 native workflow guide for the R2V input and wiring details, and its reference-to-video workflow section for the specific frame-rate rule.
Run a focused acceptance check
Use a small reference set first if you are still checking your workflow mapping. The table below is a hypothetical example, not a report of software testing or a guaranteed conversion result.
| Check | Hypothetical observation | Decision |
|---|---|---|
| Source and converted rate | Source reports 30 fps; converted reference reports 24 fps | Use the converted file for the R2V reference |
| Video order | The intended motion clip is ref_video_0 | Record index 0 in the worksheet |
| Accompanying soundtrack | Its audio is connected to ref_video_audio_0 | The video and its soundtrack indices match |
| Standalone audio | A separate voice sample is connected as the next audio reference | Record it separately; do not shift the video pairing |
| Final timeline | The editing project is set to 30 fps | Leave this as a separate delivery decision |
Listen and inspect the converted file in your chosen media tool if timing or continuity matters. The ComfyUI reference guide establishes the input interpretation and wiring rules; it does not establish that a conversion preserves every source frame, synchronizes audio, or meets your delivery timeline.
If the result is still wrong
If motion seems unexpectedly fast, slow, or sparse, first recheck the converted file’s reported rate and confirm that the workflow loads that file rather than the original. Then inspect the ref_video_N order and whether the audio was connected through the matching ref_video_audio_N input. These checks address the documented input contract; they do not prove that a particular conversion method caused or fixed a visual artifact.
If the image reference is mapped correctly but the sound is absent, check whether its video soundtrack was actually connected as a separate reference and whether the required VAE inputs are wired. If speech content matters, note a further limitation: the guide says audio references carry timbre, not words. State the intended words or delivery in the prompt rather than assuming the reference audio supplies them.
Converting a reference does not set the output video’s frame rate. Choose and verify the final timeline or export rate in the downstream editing workflow. The H3 guide’s 24 fps statement concerns how R2V reads reference video.
Workflow boundary and sources
This is documentation-based guidance with constructed examples, not a reported generation or application test. Use Imagild Film Script for a written plan and inspect the actual supported input roles in Film Prompt before reference-video recreation. External software controls belong to their named products. No exact reconstruction, native analyzer, existing-clip repair or ranking result is promised.
- ComfyUI MiniMax H3 Native Workflows — R2V reads video at 24 fps, with separate matching soundtrack inputs and modality ordering.
- Runway Image to Video Prompting Guide — Motion planning and source/temporal review principle
Editorial revision: November 13, 2026.