MiniMax-H3-Tape-FX

MiniMax-H3-Tape-FX

One node. Three dead formats. Infinite damage.

MiniMax-H3-Tape-FX is a single mega-node for ComfyUI that degrades your images and video batches with authentic analog-tape playback artifacts: VHS, BetaMax and LaserDisc — including the worn out tape look, tracking errors, dropout, creases, ghosting, head-switch noise, vertical roll and a period-correct VCR on-screen display.

Left to right: original, VHS (worn out, with OSD), BetaMax (lightly worn), LaserDisc (worn out).

Highlights

• Format DNA — each format is simulated with its real technical character, not a single filter with different names
• Six transport damage effects — tracking error bands, signal dropout, tape creases, frame ghosting, head-switch noise bar, vertical hold roll
• Worn-section simulation — damage spikes inside a configurable frame range, exactly like a bad patch of tape
• Damage presets + full manual control — mint through destroyed, every slider always available
• Authentic VCR OSD — PLAY ▶ / blinking REC ●, tape counter and date stamp, drawn with a built-in 5x7 pixel font (no font files needed)
• Video-ready — the IMAGE batch dimension is treated as a timeline, so temporal effects (tracking, rolls, ghost trails, counter) animate correctly on AnimateDiff / video latent outputs
• Deterministic — same seed, same damage, on every machine
• Zero dependencies — pure PyTorch + NumPy, both ship with ComfyUI

Demo

Worn-section simulation (animated) — the tape plays fine, then hits the damaged patch at ~0:01, then recovers:

Installation

1. Inside your ComfyUI folder: cd custom_nodes/
2. Either clone this repo or copy the MiniMax-H3-Tape-FX folder there:

   git clone https://github.com/your-name/MiniMax-H3-Tape-FX
   

3. Restart ComfyUI. Done — no extra requirements.

Quick start

1. Add the node: double-click the canvas and search for MiniMax-H3-Tape-FX (category MiniMax-H3-Tape-FX).
2. Wire any IMAGE through it — a Load Image, your sampler output, or an AnimateDiff / video batch.
3. Pick a format and a preset, then fine-tune the sliders.
4. For video: feed the whole batch at once. Set osd_fps to your frame rate so the counter and REC blink stay in sync.

An example workflow (Load Image → MiniMax-H3-Tape-FX → Save Image) is in examples/minimax_h3_tape_fx_workflow.json — drop it onto the ComfyUI canvas.

The three formats

Even a mint VHS stays soft and bleeds a little — that's the format's bandwidth, not damage. LaserDisc never creases and has no head-switch bar, but old discs rot: sparkle dropout and crosstalk ghosts instead.

Presets

Presets are a global damage multiplier applied on top of the damage sliders (mint 0.08x, lightly_worn 0.4x, worn_out 1.0x, destroyed 1.9x, custom 1.0x). The look sliders (softness, bleed, scanlines) are never scaled by presets — a mint VHS is still a soft VHS.

• mint — clean playback, format character only
• lightly_worn — occasional noise, rare dropout
• worn_out — the classic rental tape (default)
• destroyed — the tape your VCR ate and you rewound with a pencil
• custom — the sliders, exactly as you set them

Worn sections

Tapes damage in places, not uniformly. Set wear_start / wear_end (frame range, 4-frame ramps) and wear_boost, and damage spikes inside that range: more tracking chaos, heavier dropout, extra noise — then the tape plays on as before. Leave wear_end <= wear_start to disable.

On-screen display (OSD)

The OSD uses a built-in 5x7 pixel font supporting A-Z 0-9 : . - / ( ) ' plus the ▶ and ● symbols (lowercase is converted to uppercase; unsupported characters become spaces).

Full parameter reference

Format & preset

Damage effects (scaled by preset + wear)

Analog look (not scaled by presets)

Worn section

OSD

See the table above.

Misc

Tips

• Single stills always show their damage: a tracking band and (for heavy settings) a crease are guaranteed on frame 0, so the effect you dialed in is visible immediately.
• Resolution independent: effect sizes scale with image resolution (calibrated to ~640px reference).
• Performance: everything runs on your ComfyUI device (GPU when available). Typical cost is a few dozen small convolutions per frame.
• Stack it: the node output is a normal IMAGE — chain two instances (e.g. BetaMax look + extra dropout) or follow it with any color/grade node.
• Performance at 4K: raise osd_scale to keep the OSD proportional.

File structure

MiniMax-H3-Tape-FX/
├── __init__.py                    # ComfyUI entry point
├── minimax_h3_tape_fx.py          # The node (pure torch + numpy)
├── examples/
│   ├── minimax_h3_tape_fx_workflow.json
│   └── demo/                      # All images in this README
└── LICENSE                        # MIT

License

MIT — see LICENSE.