The main one, H3 Studio, builds the whole prompt and conditioning, and runs a whole project of clips: a timeline, links between clips, a song across all of them, and one Render button. Around it: a node to watch the shot while it samples, one to write the file, one to join a continuation onto the clip it continued from, and one to lock real audio into the latent.

I make short films with these. Everything here exists because I needed it, and the defaults are what I actually use.

<a href="https://youtu.be/aEUM6sihSdI"> <img width="470" height="758" alt="Capture d'écran 2026-10-17 164418" src="https://github.com/user-attachments/assets/55793f48-b879-4a99-82ff-ce2ff1c14918" /> </a> THE RECITATION — a short film made with these nodes <a href="https://youtu.be/aEUM6sihSdI"> <img width="500" height="750" alt="Capture d'écran 2026-08-17 221233" src="https://github.com/user-attachments/assets/d9984423-f965-470b-80c8-3faebffa079b" /> </a> THE RECITATION — a short film made with these nodes

What's new 😎

Latent hold — chained clips stay clean. Six links in a row and the colour, contrast and detail stay flat, where the old way went duller and crunchier at every join. More below.
Project timeline with links — 🔗 between two clips means "this one continues that one". One Render button, pick what to render.
Song mode — one track over the whole project. Each clip locks its own window of it, so H3 animates against the real song. Lay out a whole chain from the song in one click.
Timeline audio — place audio in the clip at a time, H3 fills the gaps. Play it, scrub it, drop files straight on it.
Glide Audio Lock — the same audio lock as a standalone node.

Install

Search for CGlide in ComfyUI Manager and pick nightly, or:

cd ComfyUI/custom_nodes
git clone https://github.com/CGlide/ComfyUI-CGlide

Restart ComfyUI. The nodes appear under CGlide in the node menu.

Updating: if you cloned it (or installed nightly), git pull inside the ComfyUI-CGlide folder, or Manager → Update. Then restart and hard-refresh the browser (Ctrl+F5).

Careful: the Manager's registry version installs into a folder called cglide, not ComfyUI-CGlide, and it's not a git clone, so git pull does nothing there. If you have a cglide folder, delete it and clone instead. Don't keep both, the pack loads twice.

Update ComfyUI while you're at it. Honestly, half the H3 problems people post about are an old ComfyUI, not the model. The low-VRAM path below needs 0.31.0 or newer.

Requirements

PyAV — reading reference videos and their soundtracks. Usually already there with ComfyUI.
torchaudio — resampling reference audio.
ffmpeg on your PATH — Glide Video shells out to it. Nothing else uses it.

Models

Everything comes from the Comfy-Org MiniMax H3 repo:

the reference-to-video checkpoint (and/or first-last-frame) → models/diffusion_models/
the text encoder → models/text_encoders/
the video VAE → models/vae/
the audio VAE → models/vae/

Less VRAM: grab the w4a8 model from Kijai's experimental repo — about 11.8 GB instead of 21 — and the int8_convrot video VAE from the same place. Same workflow, same prompts, you only swap the loaders.

Careful, this one is nasty: if your ComfyUI is too old for w4a8 you get a black video with sound. No error, nothing. If you see that, it's the version, not you.

H3 Studio

One node instead of the pile of loaders and text boxes. Nine image slots, three video, three audio, the prompt, the checks, and the whole project — all in one panel.

It does not touch sampling. Same sampler, same steps as a normal H3 workflow. It builds the prompt and the conditioning, that's it.

Inputs: clip, vae, audio_vae, and first_frame / last_frame for the first-last-frame family.

Outputs: positive, latent, width, height, length, seconds, overlap_frames, source_video, guide_frames, positive_refine, clip_name, locked_audio.

overlap_frames, source_video and locked_audio go to Glide Join, see below. positive_refine is for a second refine pass after a latent upscale (same prompt, no continuation anchor, bigger references if you set refine above 1×). Everything else you wire the way you would with the native H3 nodes.

What's in the panel

Both model families in one node. First / last and Omni references is a switch at the top; the accent colour follows so you always know which one you're in.
Drag files onto slots. Or paste — Ctrl+V drops a screenshot straight into the slot under your pointer. Type @ in the prompt and pick a reference to drop its tag in.
Several images at once. The nine image slots take a whole batch, from the picker or from a drop. Drop onto a particular card and the batch fills from that card forward; drop onto the rack and it fills whatever slots are empty and leaves the filled ones alone.
Drag a picture onto another slot to swap them. The @tags travel with the picture — move @image1 to slot 3 and the prompt is rewritten to @image3, so every sentence you wrote about it still describes it. Hold Alt while releasing to move the picture and leave the prompt alone.
"Sent to the encoder as". The model sees Picture 1, Picture 2 and so on, and the number comes from which slots are filled, not from the slot number. Leave a gap and your Picture 3 is not the one you think it is. This strip shows you what actually goes out.
Trim bars with playback. Drag anywhere on the track to slide the window and the picture follows, so you can hunt for the right moment.
Prompt check. Amber when the soundscape doesn't cover the whole clip, or when nothing says what the mouths are doing. Those two are most of the mumbling problems.
Shot timeline. Type a shot marker and the bar splits. One colour per shot, the first words of each shown in the band, and dragging a boundary rewrites the timestamp in the prompt. Grey means nothing is scripted there — and unscripted time is exactly where H3 invents sound.

Projects — the timeline

Several clips in one file, autosaved. Open the Project panel (the camera button) to see them all; the timeline strip at the top of the node shows them in order with their lengths. Click a block to switch clip. A new clip inherits your references so you're not reloading the same five images every time. Save packed writes a .h3proj.zip with the media inside — my whole twelve-clip film is about 42 MB.

🔗 Links

The little 🔗 between two blocks on the timeline is the whole idea:

Linked — the second clip continues the first. When it renders, it picks up the tail of the first one's render through CONTINUE FROM, motion and sound. One continuous take, longer than H3 gives you in one go.
Not linked — it's a cut. The clip starts fresh.

Click the 🔗 to toggle it. There's no mode to pick anywhere else, the links are the only place that decides.

Render

At the bottom of the Project panel, pick what to render and hit Render:

this clip
this chain — the linked run this clip is part of
from this clip on
whole project

Linked clips continue from each other and Glide Join stitches as it goes, so the last file of a chain is the whole chain. No separate assembly step.

+ look (on by default) — on every unlinked clip, hand it the clearest window of the clip before as a video reference. Separate shots, same room, same light. It takes video slot 1 on those clips.

While it runs the node is locked and shows which clip it's on. Stop finishes the clip it's rendering and stops there — everything already done stays done. If a render fails it stops, tells you which clip, and keeps the finished ones.

If you don't like the result, Revert in the Project panel puts the project back exactly as it was before the run.

Skipping a clip

Each row in the project list has a dot. Click it and that clip drops out of the run — the row dims, and Render passes over it. The clip stays in the project. Click again to bring it back. For when one shot in the middle needs another pass and you don't want to re-render the ten that already work.

Latent hold — chains that don't cook

This one I'm proud of.

Chaining used to work like this: decode the previous clip to pixels, take its last 22 frames, encode them again, hand them to H3. Every link did that round trip, and every round trip cooked the picture a little more. I ran the same jazz shot over six linked clips: by the end it was 14% less saturated and 55% crunchier on the edges. Bricks with halos, haze turned blotchy.

My first idea was to pull the colour back toward the reference. Didn't help at all. The problem wasn't the colour, it was the round trip.

So now H3 Studio continues from the previous clip's own latent — the real thing the sampler made, never decoded and re-encoded. Same six links: brightness, contrast, saturation and detail flat to the third decimal. There's still a tiny bit of drift from the model itself, but the picture stays clean.

Setup — one wire:

sampler ──output (LATENT)──> Glide Video  (the new "latent" input)

That's it. Glide Video then saves a .h3latent file next to every video it writes, and the next linked clip finds it by itself.

Latent hold is on by default. The checkbox is on the CONTINUE FROM card.
No .h3latent next to the clip? It quietly falls back to the old guide, nothing breaks. Re-render the first clip of a chain once with the wire plugged in and you're set.
Take the latent from the first sampler. If you upscale and refine after, that second latent is a different size and won't be used.
Check it worked: the console says latent hold: 7 latent frames (22 pictures) from ..., or why it fell back.

Cleaning up: each .h3latent is about 30 MB, and you only need it to continue from that clip. Once a chain is finished: ⚙ Settings → Delete latents… It tells you how many and how big first, and it never touches the videos.

Song mode

One song for the whole project. Each clip locks its own window of it into the audio half of the latent, so H3 animates against the real song — the drummer hits on the beat — instead of making up its own music.

Setup:

Project panel → Song → Add song… (or drop an audio file on it).
from — where in the song the project starts. hold — how hard it's locked (1 = exactly).
Want the whole song covered? Pick a clip length and hit Lay out clips. It builds one linked chain long enough for the song.
Render whole project.

The window each clip gets follows the links — a linked clip gives its 22 overlap frames back, so the song lands on the same frames as the picture across every join. lock off keeps the song in the project but renders without it. × Remove song takes it out.

Wire locked_audio from H3 Studio into Glide Join: it splices the original song samples back in, so the final file has the real track, not the VAE's copy of it.

Timeline audio

The three audio slots have two modes, switch at the top of the audio section:

Reference — H3 hears them and you point at them with @audio in the prompt. The usual way: a voice, a sound to copy.
Timeline — they're placed in the clip. Each one is held at the time you set, untouched, and H3 generates its own sound around it. A line of dialogue at 3 s, a door slam at the end, two songs with H3 singing the bit in between — I tried that last one, it transitions on its own. No @audio tokens in this mode, describe the sound in words.

The lane shows the clip's own time, with each audio as a coloured block:

Drop an audio file straight on the lane. It goes into the first empty slot, starting where you let go.
Drag a block to move it, drag an edge to trim it. Arrow keys nudge a frame, Shift a second.
A long file is trimmed to fit the clip, so the edges move right away.
at / from / to / len / hold on each card if you want exact numbers.

▶ Play on the left of the lane plays what this clip will hold: its part of the song with the placed audio on top. Click or drag the ruler to move the red playhead. When it reaches the end it goes back to where you pressed play, so you can loop a section.

Glide Audio Lock

The same audio lock as a standalone node, for your own workflows: it pins a real track into the audio half of an H3 latent so H3 generates the picture against it instead of re-performing it.

Glide Preview

ComfyUI's preview shows you one frozen latent frame while it samples. So you sit there for ten minutes looking at a still and you have no idea if the motion is working.

Drop this between your model and the sampler and you watch the shot move as it denoises.

mode — animated, contact sheet, or stock.
decoder — latent2rgb is free but mushy. Put taeh3.safetensors (from madebyollin) in ComfyUI/models/vae_approx/ and pick it here for real picture. Restart before it shows in the list.
every_n_steps — raise it if previewing is costing you sampling time.
max_resolution — costs decode time as well as bandwidth.

Honest limit: it caps around 7.2 fps and I can't fix that. H3 stores about one latent frame per 3.35 output frames, so that's every frame there is. It's a preview, not the render — but it's enough to kill a bad take at step four instead of at minute ten. If the decoder fails it falls back to the stock preview and your run keeps going.

Glide Video

Frames and audio into one file, in one ffmpeg pass, no second file to mux.

Presets from very compatible up to 4:4:4 10-bit and lossless FFV1. The 4:4:4 10-bit ones are visibly better on H3 output — less flat colour, more nuance. Read the tooltip, each preset says what it's for.
fps is playback rate only. Setting it wrong changes speed, it does not add or drop frames. H3 is 24.
filename_prefix — a slash makes a subfolder (jazz/clip), and date placeholders work: %date:yyyy-MM-dd%. Thanks Muxelmann for that one 👍
save_metadata embeds the prompt and workflow in the file comment. Drag-drop restore into ComfyUI is reliable for MKV, hit or miss for MP4.
fallback_on_failure retries with a more compatible preset instead of throwing away a finished render. It says so loudly when it does.
latent (optional) — the first sampler's latent, saved next to the video as a .h3latent for latent hold.

The 4:4:4 and ProRes presets don't play in a browser, so the node writes a small playable copy next to the real file just for the preview. The real file is the one in your output folder — the copy is only so you can watch it in the node.

Glide Join — extending a clip

H3 gives you a few seconds at a time. This is how you get more.

H3 Studio's CONTINUE FROM slot holds the tail of your previous clip, so the model continues the motion and the sound instead of starting fresh. Glide Join then assembles the result onto the source. In a project with links, Render fills CONTINUE FROM for you.

Wiring:

H3 Studio ──positive──> sampler ──> VAE Decode ──images──> Glide Join ──> Glide Video
          ──overlap_frames──────────────────────────────>
          ──source_video────────────────────────────────>
          ──locked_audio────────────────────────────────>
                                    VAE Decode Audio ──audio──>
          sampler ──output (LATENT)───────────────────────────────────> Glide Video

Wiring overlap_frames and source_video across means the window length is typed in one place. Change the window in the node and the join follows.

Widgets:

seam_mode — early_cut keeps the continuation's version of the overlap and colours it onto the source. One continuous model trajectory, so motion carries best. early_scurve is the same join ramped over a few frames, for when the speed change at the join is visible. hard_cut keeps the source's own frames instead.
seam_blend_frames — ramp length, only read by early_scurve.
match_levels — fits gain and offset per frame against the source across the overlap. Leave it on. With latent hold it has almost nothing left to do, which is the point.
match_tail — H3 rounds its audio grid up, so every clip carries about 8 ms more sound than picture. That grows at every join. Leave it on.

Leave the CONTINUE FROM slot empty and the whole thing is a normal generation — H3 Studio sends nothing, Glide Join passes the frames straight through. One workflow for both, no rewiring, no bypassing.

The window is locked to the tail of the clip, on purpose. A window from the middle produces material that follows on from the middle, and splicing that after the last frame skips everything between — it reads as a hard cut and no seam mode can fix it.

The chroma badge

This one matters when latent hold falls back to the guide (no .h3latent next to the clip).

There's a little badge on the continuation waveform. Green for 4:4:4 and 4:2:2, orange for 4:2:0. It's not decoration.

I had a continuation that kept coming back slightly out of sync — Glide Join kept printing frame correspondence was off by -2. I went looking at the anchor window, the frame counts, the way the tail is read. Wasted a good while in there. It was the codec: AV1, which is 4:2:0. Same pair again as H.264 4:4:4 and the line just disappeared. ProRes 4:2:2, also clean.

So: don't chain from a 4:2:0 file. The badge tells you before you render. A look reference carries no timing, so there's no badge on a reference video slot.

Careful with this

A .py change needs a full ComfyUI restart, not a browser refresh.
If a node's widget list changes shape between versions, delete and re-add the node. ComfyUI matches saved widget values by position and will smear old values across new widgets.
Long chains still drift a little, even with latent hold — it's the model pushing its own look at every link. Fine for a few links; for a long one-take, grade it in post or cut and start a fresh chain.
There will be bugs — open an issue and tell me what you did, that's genuinely more useful than a star.

Credits

MiniMax for H3, and the ComfyUI team for the core H3 nodes this builds on.
madebyollin for the TAE decoder that makes the preview readable.
Kijai for the quantised models that put this on smaller cards.
Muxelmann for the date placeholders in Glide Video — first outside contribution to the pack 🙏

Development assistance from Claude (Anthropic).