Hello, I've ran your previous WAN models that were larger than my VRAM without issue, so as a pretense know my VRAM is about 12GB with 16GB RAM and i'm not sure if LTX2 is incompatible with lowvram or not, kinda feels like it doesn't like that setting.

The blunt description of the issue: upon getting to the UNET config/loader, it starts to load, then python just disengages, comfy shuts down, no warning/error, no reason, no log, just goes to OS CLI silent.

I've ran GGUF versions with no problem on the UNET loader at least, and have had success with configurations, on your workflow no less, with GGUF versions.

I have then had the idea to attempt to convert to GGUF your model.

However, there seems to be some tensor related issues with the base of the model, preventing the GGUF conversion (after converting to fp8.safetensors -> F16.GGUF with success, though, with a relaxed max of 127 vs 64.)

I can't help but to wonder if my stability issues are related to this tensor size being exceeded to begin with, either way, if you want a productive relationship with anyone converting your model to GGUF these would be best to be addressed, or for your own conversion attempts, etc.

below is the error I'm getting and my analysis results in raw form:

ERROR:
llama-quantize.exe "ltx-2-19b-phr00tmerge-nsfw-v6-F16.gguf" "ltx-2-19b-phr00tmerge-nsfw-v6-Q4_K_M.gguf" Q4_K_M
main: build = 7772 (287a33017)
main: built with Clang 19.1.5 for Windows x86_64
main: quantizing 'ltx-2-19b-phr00tmerge-nsfw-v6-F16.gguf' to 'ltx-2-19b-phr00tmerge-nsfw-v6-Q4_K_M.gguf' as Q4_K_M
gguf_init_from_file_impl: tensor name 22 is too long: 65 >= 64
gguf_init_from_file_impl: failed to read tensor info
llama_model_quantize: failed to quantize: llama_model_loader: failed to load model from ltx-2-19b-phr00tmerge-nsfw-v6-F16.gguf
main: failed to quantize model from 'ltx-2-19b-phr00tmerge-nsfw-v6-F16.gguf'

Analysis(Results):

PROBLEMATIC TENSORS (name length >= 64):

Tensor 22: 65 chars -> av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_1.weight
Tensor 24: 65 chars -> av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_2.weight
Tensor 27: 72 chars -> av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_1.bias
Tensor 28: 74 chars -> av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_1.weight
Tensor 29: 72 chars -> av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_2.bias
Tensor 30: 74 chars -> av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_2.weight
Tensor 34: 65 chars -> av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_1.weight
Tensor 36: 65 chars -> av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_2.weight
Tensor 39: 72 chars -> av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_1.bias
Tensor 40: 74 chars -> av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_1.weight
Tensor 41: 72 chars -> av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_2.bias
Tensor 42: 74 chars -> av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_2.weight

Analysis(Raw form from list):

15 40 audio_caption_projection.linear_2.weight
16 24 audio_patchify_proj.bias
17 26 audio_patchify_proj.weight
18 19 audio_proj_out.bias
19 21 audio_proj_out.weight
20 23 audio_scale_shift_table
21 63 av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_1.bias
22 65 av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_1.weight <-- TOO LONG!
23 63 av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_2.bias
24 65 av_ca_a2v_gate_adaln_single.emb.timestep_embedder.linear_2.weight <-- TOO LONG!
25 39 av_ca_a2v_gate_adaln_single.linear.bias
26 41 av_ca_a2v_gate_adaln_single.linear.weight
27 72 av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_1.bias <-- TOO LONG!
28 74 av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_1.weight <-- TOO LONG!
29 72 av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_2.bias <-- TOO LONG!
30 74 av_ca_audio_scale_shift_adaln_single.emb.timestep_embedder.linear_2.weight <-- TOO LONG!
31 48 av_ca_audio_scale_shift_adaln_single.linear.bias
32 50 av_ca_audio_scale_shift_adaln_single.linear.weight
33 63 av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_1.bias
34 65 av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_1.weight <-- TOO LONG!
35 63 av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_2.bias
36 65 av_ca_v2a_gate_adaln_single.emb.timestep_embedder.linear_2.weight <-- TOO LONG!
37 39 av_ca_v2a_gate_adaln_single.linear.bias
38 41 av_ca_v2a_gate_adaln_single.linear.weight
39 72 av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_1.bias <-- TOO LONG!
40 74 av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_1.weight <-- TOO LONG!
41 72 av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_2.bias <-- TOO LONG!
42 74 av_ca_video_scale_shift_adaln_single.emb.timestep_embedder.linear_2.weight <-- TOO LONG!
43 48 av_ca_video_scale_shift_adaln_single.linear.bias
44 50 av_ca_video_scale_shift_adaln_single.linear.weight
45 32 caption_projection.linear_1.bias
46 34 caption_projection.linear_1.weight
47 32 caption_projection.linear_2.bias
48 34 caption_projection.linear_2.weight
49 18 patchify_proj.bias
50 20 patchify_proj.weight
51 13 proj_out.bias
52 15 proj_out.weight
53 17 scale_shift_table
54 40 transformer_blocks.0.attn1.k_norm.weight

I will attempt one bypass or another, seeing about increasing the quantize size limit comes to mind first.
if that fails may try either removing the problem tensors/neutralizing the slot, or concat/compress them in some way.
after either one if needed I may look into a custom lama patch if everything doesnt go over 100% smooth

Though I do suggest getting these tensors within compliance, as mentioned, for future version stability and compatibility.