ComfyUI Node
Train Anima Slider LoRA
Train an experimental Anima/Cosmos flow-slider LoRA from a prompt YAML using the connected MODEL and CLIP.
Train Anima Slider LoRA
- model
- clip
- vae
- lora
- report_json
- lora_path
- report_path
◄prompt_yamlprompts-anima-breast_size_slider.yaml►
◄custom_prompt_yaml_path►
◄prompt_indices0,1,2,3,4,5►
◄eval_prompt_indices6,7►
◄steps600►
◄lr0►
◄rank16►
◄alpha16.0►
◄network_presetattn_mlp►
◄network_reg_dims►
◄network_reg_lrs►
◄model_residencyprefer_cuda►
◄lora_weight_dtypefp32►
◄gradient_checkpointingtrue►
◄skip_initial_evaltrue►
◄skip_final_evalfalse►
◄width0►
◄height0►
◄num_inference_steps20►
◄timestep_samplingshift►
◄sigmoid_scale1.00►
◄discrete_flow_shift3.00►
◄loss_weighting_schemenone►
◄direction_lossenhance_only►
◄teacher_guidance_scale1.00►
◄teacher_norm_referencepositive►
◄min_step_index-1►
◄max_step_index-1►
◄eval_step_indices►
◄eta1.00►
◄seed961218314523996►
◄eval_seed961218314523996►
◄vary_seedtrue►
◄allow_unsafe_age_termsfalse►
◄output_lora_prefixloras/anima_slider►
Categorytraining/anima slider
Inputs (38)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Loaded diffusion model to train against. | |
| clip | CLIP | Loaded text encoder used to encode the prompt YAML. | |
| vae | VAE | Accepted for workflow parity with Anima pipelines; the current text-only trainer does not encode images. | |
| prompt_yaml | COMBO | prompts-anima-breast_size_slider.yaml | Prompt YAML bundled with this custom node. |
| custom_prompt_yaml_path | STRING | Optional absolute or ComfyUI-working-directory-relative YAML path. Overrides prompt_yaml when set. | |
| prompt_indices | STRING | 0,1,2,3,4,5 | Comma-separated prompt indices to cycle during training. |
| eval_prompt_indices | STRING | 6,7 | Comma-separated prompt indices for before/after eval. Empty uses prompt_indices. |
| steps | INT | 6001–100000 | Training optimizer steps. |
| lr | FLOAT | 00–1 | Fallback LoRA learning rate. |
| rank | INT | 161–256 | Fallback LoRA rank. |
| alpha | FLOAT | 16.00–1024 | LoRA alpha. |
| network_preset | COMBO | attn_mlp | LoRA target preset. |
| network_reg_dims | STRING | Optional YAML mapping of regex fullmatch patterns to LoRA ranks. | |
| network_reg_lrs | STRING | Optional YAML mapping of regex fullmatch patterns to learning rates. | |
| model_residency | COMBO | prefer_cuda | Best-effort base model residency after ComfyUI loading. Falls back to DynamicVRAM behavior if CUDA promotion fails. |
| lora_weight_dtype | COMBO | fp32 | Trainable LoRA weight dtype. fp32 is recommended; base/bf16 reduce VRAM but may lose small updates. |
| gradient_checkpointing | BOOLEAN | true | Checkpoint trainable diffusion blocks during LoRA training to reduce activation VRAM. |
| skip_initial_eval | BOOLEAN | true | Skip the pre-training eval pass for OOM isolation. Not a quality substitute. |
| skip_final_eval | BOOLEAN | false | Skip the post-training eval pass for OOM isolation. Not a quality substitute. |
| width | INT | 00–4096 | Training latent width in pixels. 0 uses the selected prompt YAML resolution. |
| height | INT | 00–4096 | Training latent height in pixels. 0 uses the selected prompt YAML resolution. |
| num_inference_steps | INT | 203–200 | Number of simple scheduler sigmas. |
| timestep_sampling | COMBO | shift | 6 options: uniform, mid, early_late, sigmoid, shift, flux_shift |
| sigmoid_scale | FLOAT | 1.000.01–20 | — |
| discrete_flow_shift | FLOAT | 3.000.01–20 | — |
| loss_weighting_scheme | COMBO | none | 3 options: none, sigma_sqrt, cosmap |
| direction_loss | COMBO | enhance_only | 2 options: enhance_only, bidirectional |
| teacher_guidance_scale | FLOAT | 1.000–20 | Global multiplier applied after each prompt YAML guidance_scale. |
| teacher_norm_reference | COMBO | positive | Output norm reference for the teacher signal. neutral usually makes stronger sliders less prone to scale blow-up. |
| min_step_index | INT | -1-1–10000 | -1 uses the default lower bound. |
| max_step_index | INT | -1-1–10000 | -1 uses the default upper bound. |
| eval_step_indices | STRING | Comma-separated eval step indices. Empty uses midpoint. | |
| eta | FLOAT | 1.000–20 | — |
| seed | INT | 9612183145239960–18446744073709550000 | — |
| eval_seed | INT | 9612183145239960–18446744073709550000 | — |
| vary_seed | BOOLEAN | true | — |
| allow_unsafe_age_terms | BOOLEAN | false | — |
| output_lora_prefix | STRING | loras/anima_slider | Output prefix under the ComfyUI output directory. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| lora | LORA_MODEL | — |
| report_json | STRING | — |
| lora_path | STRING | — |
| report_path | STRING | — |