GGUF version of the original LTX-2 Rapid, converted in colab (this is link to the jupyter notebook script) using city96 gguf convertion instructions with manual adding config field to metadata.

Roadmap

Quants with fixed config
GGUF version of workflow
Demo videos for different quants on the same noise seed

If someone will provide a test result for different quants to compare, it will be too good! PR into this repo

Developing

If you're making custom ggufs, do not forget to add this to gguf metadata into the config string field:

	{"transformer": {"_class_name": "AVTransformer3DModel", "_diffusers_version": "0.25.1", "activation_fn": "gelu-approximate", "attention_bias": true, "attention_head_dim": 128, "attention_type": "default", "caption_channels": 3840, "cross_attention_dim": 4096, "double_self_attention": false, "dropout": 0.0, "in_channels": 128, "norm_elementwise_affine": false, "norm_eps": 1e-06, "norm_num_groups": 32, "num_attention_heads": 32, "num_embeds_ada_norm": 1000, "num_layers": 48, "num_vector_embeds": null, "only_cross_attention": false, "cross_attention_norm": true, "out_channels": 128, "upcast_attention": false, "use_linear_projection": false, "qk_norm": "rms_norm", "standardization_norm": "rms_norm", "positional_embedding_type": "rope", "positional_embedding_theta": 10000.0, "positional_embedding_max_pos": [20, 2048, 2048], "timestep_scale_multiplier": 1000, "av_ca_timestep_scale_multiplier": 1000.0, "causal_temporal_positioning": true, "audio_num_attention_heads": 32, "audio_attention_head_dim": 64, "use_audio_video_cross_attention": true, "share_ff": false, "audio_out_channels": 128, "audio_cross_attention_dim": 2048, "audio_positional_embedding_max_pos": [20], "av_cross_ada_norm": true, "use_embeddings_connector": true, "connector_attention_head_dim": 128, "connector_num_attention_heads": 30, "connector_num_layers": 2, "connector_positional_embedding_max_pos": [4096], "connector_num_learnable_registers": 128, "connector_norm_output": true, "use_middle_indices_grid": true, "rope_type": "split", "frequencies_precision": "float64"}, "scheduler": {"_class_name": "RectifiedFlowScheduler", "_diffusers_version": "0.25.1", "num_train_timesteps": 1000, "shifting": null, "base_resolution": null, "sampler": "LinearQuadratic"}}

(config can change and should be taken from original to work properly)

3ndetz/LTX2-Rapid-Merges-GGUF

LTX-2 Rapid GGUF

Roadmap

Developing

3ndetz/
LTX2-Rapid-Merges-GGUF