Loading
Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
Bernini-R Workflow · ComfyUI-BerniniR Wrapper
Profil d'exécution
Description de la source
Bernini-R Workflow · ComfyUI-BerniniR Wrapper
A node pack for the Bernini-R (1.3B / 14B) video generation model — VRAM optimization · multi-mode guidance · context window · differential-diffusion masking · segmented prompt control github: xiaolibai-sys/ComfyUI-BerniniRWrapper
Overview
This workflow is built on the ComfyUI-BerniniR Wrapper node pack. A single graph combines VRAM optimization, a multi-mode guidance family, temporal context windows, differential-diffusion masking, and segmented prompt control — letting long videos run on consumer GPUs / low-RAM machines while still allowing fine-grained per-segment content control.
1 · VRAM & Loading Optimization
● Block Swap Per-layer DiT weight swapping between GPU and CPU with pin_memory and asynchronous prefetch, drastically cutting peak VRAM so the 14B model runs on limited memory.
Commentaire généré par l’IA
Explication générée par l’IA à partir des détails de la source et de la configuration. Les suggestions sont clairement signalées.
This ComfyUI workflow creates image-to-video clips from an image and text prompts, with prompts assignable to frame ranges and masks for selected regions; it outputs video.
Choose among seven guidance modes—CFG, APG, RAAG, S2, Z2, STG_A, and STG_R—with cosine, linear, or piecewise per-step strength schedules.
For long videos, the workflow pairs FreeNoise noise shuffling with overlapping temporal windows to support motion and semantic coherence across segments.
A dual expert sampler switches between high-noise and low-noise models at a selected step; NAG adds a negative-prompt attention path.
Enter segmented prompts as `1-32(index1): prompt1 ; 33-64(index2): prompt2`; each index maps to a window in BerniniR_SegmentSchedule.
Differential-diffusion masks can freeze or anneal selected regions for inpainting, reference-image preservation, or regional conditioning.
The workflow is built on the ComfyUI-BerniniR Wrapper and uses the Wan Video 2.2 I2V-A14B base model.
The workflow file lists these model files: `Bernini-R_LightX2V_high_noise.safetensors`, `Bernini-R_LightX2V_low_noise.safetensors`, `wan_2.2_vae.safetensors`, `wan_umt5-xxl_bf16_fixed.safetensors`, `wan2.2_bernini_low_noise_fp8_scaled.safetensors`, and `Wan22_Bernini_HIGH_fp8_e4m
Automatic windowing means you do not need to configure a Context Window manually.
Tip: Choose Full (eager) loading when RAM is ample. On memory-constrained systems, streaming loads weights from disk on demand and uses less peak memory, but starts more slowly.
Suggestion · non vérifié
GGUF quantized weights are not supported; the workflow loads standard safetensors checkpoints.
For storyboard or multi-scene shots with large semantic jumps, consider an explicit Context Window or separate batches; shared local context is weaker across those transitions.
Suggestion · non vérifié
Faut-il modifier cela ?
Connectez-vous pour envoyer une demande de modification.
Sources
1 sourceExtraits de sources
2 extraitsSource context: 164 downloads · Type Workflows · Base model Wan Video 2.2 I2V-A14B
● Streaming Loader Optimized for low-RAM devices — weights are streamed from disk on demand instead of loading the whole model into memory at once.
● Automatic Model Offload Lazy-loading strategy: weights are loaded only when used and released when idle, avoiding two copies of the same model weights in memory and keeping the memory peak low.
2 · Guidance
A seven-mode guidance family is included: CFG, APG, RAAG, S2, Z2, STG_A, STG_R — covering classifier-free, projected, ratio-aware, stochastic self-, zero-cost zigzag, and spatiotemporal skip guidance. A per-step guidance strength schedule (cosine / linear / piecewise) is also supported. See the separate guidance documentation for details.
3 · Context & Noise Control
● Context Window Paired with FreeNoise noise shuffling, long videos are split into overlapping temporal windows to keep motion and semantics coherent across segments and avoid long-sequence degradation.
4 · Conditioning & Masking
● Differential Diffusion Mask A differential-diffusion mask mechanism supporting freeze / anneal on selected regions — useful for inpainting, reference-image preservation, or regional conditioning.
● NAG (Normalized Attention Guidance) Injects a negative-prompt attention path to enhance detail.
5 · Sampler
● Dual Expert Sampler Switches between high-noise and low-noise expert models by split step, balancing early structure formation with late detail convergence.
6 · Segmented Prompt Control
● BerniniR SegmentSchedule Generates differing prompts between windows for per-segment semantic choreography.
Prompt Format
1-32(index1): prompt1 ; 33-64(index2): prompt2
Each segment is tagged with a frame range + index, separated by semicolons; the index maps to a window in SegmentSchedule.
Recommendations
● Auto windowing The workflow windows automatically — no manual Context Window configuration needed.
● Recommended: continuous motion control Windows share local context, ideal for character action, camera moves, and other strongly continuous tasks.
● Use Full load when RAM is ample If your machine has enough RAM, prefer the Full (eager) load mode over the streaming loader — weights stay resident in memory so startup and generation begin faster; the streaming loader only pays off on memory-constrained systems.
Note: storyboard / multi-scene shots perform worse — because local context is shared, large semantic jumps between segments are weak. For those, use an explicit Context Window or separate batches.
Limitations
● No GGUF support The pack loads standard safetensors checkpoints; GGUF quantized weights are not supported, so GGUF-based quantized / CPU-offload paths cannot be used.
● Streaming loader is slower Reading weights from disk on demand makes initial load and generation startup slower than eagerly loading the full model into RAM. It trades speed for a lower peak memory footprint on low-RAM machines.
Remaining hardware or download-size estimates are not available yet.
Exigences
Bernini-R_LightX2V_high_noise.safetensors
Fichier non vérifiéCheckpoint · Hugging Face · rzgar/Bernini-R-LightX2V-4step-loras · Bernini R LightX2V 4step Loras
Bernini-R_LightX2V_low_noise.safetensors
Fichier non vérifiéSource context: 48 downloads · Type Workflows · Base model Wan Video 2.2 I2V-A14B
Checkpoint · Hugging Face · rzgar/Bernini-R-LightX2V-4step-loras · Bernini R LightX2V 4step Loras
VAE · 1.31 GB · SAFETENSORS · fp8 · pruned · Civitai · 1817671
wan_umt5-xxl_bf16_fixed.safetensors
Non résoluEmbedding · Unknown
wan2.2_bernini_low_noise_fp8_scaled.safetensors
Non résoluCheckpoint · Unknown
Checkpoint · Hugging Face
