🅛🅣🅧 Dynamic Conditioning

LTX has a long-standing image-to-video weakness: it drifts. Feed it a starting frame and the output either barely moves (a slow zoom on a frozen image) or wanders off and stops resembling your input within a few frames. DynamicConditioning is a patch node aimed right at that tension - it lets you push the conditioning harder, and crucially, focus that push on the first frame, so your anchor image holds without over-constraining the rest of the clip into stillness.

That "follow the prompt vs. don't fall apart" balance is the exact axis LTX has struggled on, and the one Lightricks kept iterating (2.3's whole pitch was reworking I2V so outputs actually move instead of freezing). This node gives you a manual lever on it.

How it works

It patches your model so the conditioning is scaled by a power curve rather than applied flat. Raising the power increases how strongly the model adheres to your conditioning. With only_first_frame on, that boosted adherence is concentrated on the initial frame - the one that needs to match your input image in I2V - while the later frames are left freer to actually animate. Turn only_first_frame off and the stronger conditioning applies across the whole clip.

The inputs and outputs that matter

• model (MODEL) - the LTX model to patch.
• power (FLOAT, default 1.3, range 1–2) - the strength of the conditioning boost. 1.0 is effectively no change; higher pushes adherence up. The default 1.3 is a gentle nudge - increase it if the first frame isn't holding, but don't slam it to 2, which tends to over-constrain and can freeze motion.
• only_first_frame (BOOLEAN, default true) - whether to concentrate the boost on the first frame. On (the default) is the I2V-anchoring use case: lock the start, let the rest move. Off spreads the effect across all frames.

Output is a patched MODEL to feed your sampler.

How to install it

Install via ComfyUI Manager as the README recommends: Ctrl+M → Install Custom Nodes → search LTXVideo → Install → restart. Or manually:

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

then restart ComfyUI. It's under the LTXV menu. No downloads of its own, but it patches the LTX model, which needs 32GB+ VRAM and 100GB+ disk per Lightricks.

Common issues & troubleshooting

Your I2V output freezes / barely moves. Counterintuitively, too much conditioning strength causes this - the model clamps to the input and won't animate. Lower power toward 1.0, and keep only_first_frame on so the boost doesn't smother motion in later frames. Frozen I2V was LTX-2's single most-reported bug; over-conditioning makes it worse, not better.

Your I2V output drifts away from the input frame. The opposite failure. Raise power (with only_first_frame on) so the start frame holds more firmly. If it still drifts hard, you may be on an older LTX-2 checkpoint - 2.3 substantially reworked I2V consistency, and no amount of conditioning tuning fully substitutes for the newer weights.

No visible effect. At power 1.0 there's essentially nothing to do; you need to move it above 1 to see anything. Also confirm the patched model output is the one wired into your sampler, not the original.

Whole clip went stiff after turning only_first_frame off. That's expected - off means the strong conditioning hits every frame, which suppresses motion throughout. For I2V, keep it on. Turn it off only if you deliberately want prompt adherence pinned across the entire sequence at the cost of movement.