Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More

Owned by Michal

Create professional AI videos and long-form stories. Learn practical workflows, explore models and share techniques with other creators.

5 contributions to Muse Collective
Photo Restoration
Just wondering if anyone has or can suggest a decent work flow for doing photo-restoration? The basic Flux2 and or Krea2 workflows for the most part do a decent job. Where I run into a glitch is with some of the older pictures where the face detail is lacking. So, face swap is the obvious choice. I've been taking from one good image to place onto the bad one, but this method of replacement does not seem to keep the head alignment the same. IE, the person is looking to the left away from the camera, but the replacement is looking at the camera. All thoughts and suggestions welcome.
1 like • 5d
I would start with a reference where the head is turned in roughly the same direction as the damaged photo. If you can mask the edit, keep the original head outline and only work on the damaged facial area. Compare the eye line and nose angle before judging sharpness; a sharper face looking the wrong way still changes the photo.
Muse H3 Director V2.1 coming this month 🥳
Hey guys, I am tirelessly testing the new Director V2.1 to get it out to you as soon as I can. It all seems fine so far but better test than sorry. Here is the walkthrough. I have also posted on YouTube, so any likes etc... on the channel would be much appreciated 👍
Muse H3 Director V2.1 coming this month 🥳
1 like • 5d
where can i download the muse director?
0 likes • 5d
@Andy Whittaker ah, ok, thanks. Not upgrading for now - just wanted to test the director as I see a potential cooperation with my project (www.lorapilot.com)
What do you save before updating ComfyUI?
For those running several custom node packs, what has actually let you recover an older working setup after an update broke it? A second install, a full folder backup, something else? Interested in what you had to restore beyond the workflow JSON.
Specific Sound synced to a specific motion
Hi Andy, I appreciate your fantastic work on this advanced nodes a lot. 1. Struggling with understanding Unified Loader in Director V2: Do I have to enable the Loras in Unified Loader AND in Director settings? 2. I use an RTX-5080 and tried mostly lowram or balanced settings. Could you recommend me an optimal setting of ggufs, clips and loras for 16GB Vram? 3. In Character-Sheet V2 I cant select lowram gguf models like MiniMax-H3-ref2va-curve-Q5_1.gguf in the Models&Loras Node, only safetensors. Is this wanted? 4. Is it possible to sync a specific sound to a specific motion, e.g. a hammersmith at work producing with every stroke an unexpected sound like somebody is hammering wood instead of steel? The Audio Refs are all tied to CharacterRefs 1-3 and to speech or lipsync. Perhaps an additional category like 'specific sound tied to a prompted motion or motion REF' could add a new dimension? Any help appreciated
0 likes • 7d
For the deliberately wrong hammer sound, I would add it in the edit once the motion is right. Put a marker on each frame where the hammer makes contact, then line up the start of each wooden hit with those markers. That lets you change the joke or its timing without regenerating a good shot.
0 likes • 5d
@Daniel Steinauer Yes, I meant in DaVinci Resolve or another video editor, after exporting the clip. Sorry, I should have made that clearer. The markers go on the editing timeline, not in the Muse prompt. Then you can move each sound effect until its attack lines up with the hammer contact.
Audio fixes?
Hi any advice on how to fix H3 audio issues like unwanted distortion noses, or harsh sounds?
1 like • 10d
If this is a sequence of extended clips, listen to each raw clip separately before the final edit. Note the first one where the rumble appears and whether it starts immediately or builds within that clip. Send Andy that clip and the clean one before it; that gives him a more useful example than the whole sequence.
1-5 of 5
Michal Vavak
1
1 point to level up
@michal-v-4085
I help AI creators troubleshoot ComfyUI, train LoRAs and build repeatable image and video workflows. Creator of LoRA Pilot. Running Diffusion Lab.

Active 18h ago
Joined Oct 31, 2026
INFJ
Slovakia