ComfyUI: Difference between revisions
m (→Video Upscale) |
m (→Video Upscale) |
||
| Line 125: | Line 125: | ||
|2026-01-30 |
|2026-01-30 |
||
|343.10s |
|343.10s |
||
|seedvr2_ema_3b_fp16 |
|'''seedvr2_ema_3b_fp16''' |
||
| |
| |
||
| |
| |
||
| Line 131: | Line 131: | ||
|2026-01-30 |
|2026-01-30 |
||
|351.37 |
|351.37 |
||
|seedvr2_ema_3b_fp8 |
|'''seedvr2_ema_3b_fp8''' |
||
| |
| |
||
| |
| |
||
| Line 144: | Line 144: | ||
|442.38 |
|442.38 |
||
|seedvr2_ema_3b_fp8 |
|seedvr2_ema_3b_fp8 |
||
|With torch compile and tile size of 768 (First run) |
|With torch compile and tile size of '''768''' (First run) |
||
| |
| |
||
|- |
|- |
||
| Line 150: | Line 150: | ||
|305.02 |
|305.02 |
||
|seedvr2_ema_3b_fp8 |
|seedvr2_ema_3b_fp8 |
||
|With torch compile and tile size of 768 (Second run) |
|With torch compile and tile size of 768 ('''Second''' run) |
||
| |
|||
|- |
|||
|2026-01-30 |
|||
|408.12 |
|||
|seedvr2_ema_3b_fp8 |
|||
|With torch compile and tile size of 768, '''swap blocks 6''' ('''First''' run) |
|||
| |
| |
||
|} |
|} |
||
Revision as of 14:49, 29 January 2026
Misc
https://www.youtube.com/@goshniiAI/videos - Upscaling
Generate Million-Megapixel Images with Z Image Turbo!
How to Train a Z-Image-Turbo LoRA with AI Toolkit
Stop Making Blurry AI Art! The "Halftone" Fix for Wan & Qwen (ComfyUI Workflow)
Insane ComfyUI Upgrade: 165 Custom Nodes That Actually Work
Different Region = Different LoRA! New ComfyUI Nodes for Area Composition
Comfyui 101: How to Create X/Y Plots for SDXL & Flux How to Use Flux Unet ( & GGUF) Models With TinyTerraNodes XY Plot
https://civitaiarchive.com https://civitasbay.org/
zer0int/CLIP-GmP-ViT-L-14/tree/main - what's the best version of Vit-L clip encoder? smooth and detail both r good
Updateing
git pull
Installing
uv pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130
uv pip install sageattention==2.2.0 --no-build-isolation
pip install -r requirements.txt
Prompts
[X : Y : N] - Use X for N steps then Y. Not sure if this works in comfyui.
Prompting Guide - FLUX.2 [klein]
https://docs.bfl.ai/guides/prompting_guide_kontext_i2i
https://danbooru.donmai.us/wiki_pages/tag_groups
https://danbooru.donmai.us/related_tag
https://danbooru.donmai.us/posts?tags=list_of_style_parodies&z=2
https://betterwaifu.com/blog/ai-pose-prompts
Artists
Deak Ferrand - Dune, Sandman
by Diane Pierce and Alena Aenami in the style of Dan Mumford,
Asaf hanuka
rodolphe wytsman impressionism
by Michal Karcz, by Arthur Sarnoff, by Marc Simonetti,
Greg Rutkowski
Syd Mead
Cyberpunk
Syd Mead - Bladerunner concept art
Josan Gonzalez
Sparth (Nicolas Bouvier)
Beeple (Mike Winkelmann) - Weird colours
Simon Stålenhag - Giant robots guy
Fantasy
Greg Rutkowski
Benchmark
Video Upscale
| 2026-01-30 | 336.80s | GAN upscaler template in ColmfyUI | for a 3 second video using the | |
| 2026-01-30 | 19.04s | GAN upscaler template in ColmfyUI | for a second pass (Cached or load time?) | |
| 2026-01-30 | 343.10s | seedvr2_ema_3b_fp16 | ||
| 2026-01-30 | 351.37 | seedvr2_ema_3b_fp8 | ||
| 2026-01-30 | 354.54 | seedvr2_ema_3b-Q4_K_M | ||
| 2026-01-30 | 442.38 | seedvr2_ema_3b_fp8 | With torch compile and tile size of 768 (First run) | |
| 2026-01-30 | 305.02 | seedvr2_ema_3b_fp8 | With torch compile and tile size of 768 (Second run) | |
| 2026-01-30 | 408.12 | seedvr2_ema_3b_fp8 | With torch compile and tile size of 768, swap blocks 6 (First run) |
Seed VR2
SeedVR2 BEST SETTINGS for Every GPU! (6GB–32GB+) Nobody Told You This…
LTX-2
LTX 2 Q6 16GB on 8GB VRAM? 🤯 Free ComfyUI Workflow 2026
LTX-2 generate a 30s video in 310seconds - Has distilled GGUFs
LTX-2 Simplified Workflow 🔥 Distilled Checkpoints or Separated VAE & Transformer?
LTX-2 Fix: Yes, You Can Actually Use It Now (16-24gb VRAM) - Didn't work
This fixed my OOM issues with LTX-2 - Didn't work.