Pinokio
FeedAppsRemoteCreate
Author@blizaine
Download Pinokio
Sign inCreate accountJoin
Sign inCreate accountJoin

@blizaine

GitHub
GitHub
SupportKo-fiGitHub Sponsors

Joined February 2026

4 Following260 Followers3 Apps
PostsRepliesApps
More
ActivityCheckpointsFollowed appsGitHub creations

Published apps

  • MaestroAn all-in-one, 100% local AI creative studio, director, and multi-track editor. Generate with MiniMax H3, LTX-2.5/2.3, Wan, Flux, Qwen, and more; turn an idea or song into a planned production; then finish it on the timeline. Requires an NVIDIA GPU (6GB+ VRAM).
  • Qwen3-TTS MLX WebUI EnhancedHigh-quality text-to-speech with Beautiful Web UI & API, optimized for Apple Silicon using MLX. Features include Custom Voice (preset speakers), Voice Design (natural language), and Voice Cloning. With enhanced features for saving custom voices and long-form / endless TTS streaming.
  • chatterbox
View all apps
Replies · 16
@corwiiSeem to be having a huge slowdown since the recent updates. Had to roll back to version 2.7.0 to gain my speeds again with Minimax H3. Gemini wrote this issue report. I can confirm that rolling back to 2.7.0 and disabling all the MMGP stuff fixed my issue and I'm back to normal speeds. Disabling MMGP on version 3.1 did
@blizaine·1mo ago
in Maestro
Can you update and run the Dev Branch? I pushed a fix for 16GB cards, but I can't test it.
@corwiiSeem to be having a huge slowdown since the recent updates. Had to roll back to version 2.7.0 to gain my speeds again with Minimax H3. Gemini wrote this issue report. I can confirm that rolling back to 2.7.0 and disabling all the MMGP stuff fixed my issue and I'm back to normal speeds. Disabling MMGP on version 3.1 did
@blizaine·1mo ago
in Maestro
Do you have the actual commit that you rolled back to that works? v2.7 had several. Also, are you using the standard pruned/full models or one of the fused?
@corwiiSeem to be having a huge slowdown since the recent updates. Had to roll back to version 2.7.0 to gain my speeds again with Minimax H3. Gemini wrote this issue report. I can confirm that rolling back to 2.7.0 and disabling all the MMGP stuff fixed my issue and I'm back to normal speeds. Disabling MMGP on version 3.1 did
@blizaine·1mo ago
in Maestro
Thanks. I’ll see what I can do.
@jayreddinHi everyone! Just wanted to say hello to the whole Pinocchio community
@blizaine·1mo ago
hey there! 👋
@cocktailpeanutHey @blizaine I think there's an overall issue with the model downloads. I heard several people reporting this issue with ViggleAnimate, but this one was H3 VDN First/Last Pruned.
@blizaine·1mo ago
in Maestro
update: Maestro has 930 huggingface URL references. 69 where broken and are now fix on public Dev. I'll be pushing to main later today. craziness.
@cocktailpeanutHey @blizaine I think there's an overall issue with the model downloads. I heard several people reporting this issue with ViggleAnimate, but this one was H3 VDN First/Last Pruned.
@blizaine·1mo ago
in Maestro
public Dev includes download fixes for Viggle Animate, Face Refiner, H3 x2 VAE, H3 VDN modules and LoRAs, Qwen 2.1 processors, and YuE2 assets. I'm having my agent audit every HF reference in Maestro. apparently there are 457. 😂😭 Dev v3.2 has a ton of fixes. I'll push to Main branch today.
@corwiiCan't get H3 VAE X2, not sure if it was mentioned yet. Error shows up when selecting it before generating. Entry Not Found for url: https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/adc81ccb71352192214d83d5fafb9487e860be39/minimax_h3/MiniMax-H3-X2-Detail-v1_int8_convrot.safetensors.
@blizaine·1mo ago
in Maestro
I think the repo was changed or moved. v3.1 will have a fix. Plan to release this evening with a some other updates.
@primeectoScrew Flux 3 klein, get MOSS image 5B in here.
@blizaine·1mo ago
in Maestro
agree! lol, it's on the public Dev branch. pushing to main later today.
@mykojagsinI get a huggingface error when I try the refine faces option. Other than that, I love this app. Great work.
@blizaine·1mo ago
in Maestro
Just pushed v3.1 to the Public Dev branch for testing. Push to Main should be coming later today if all goes well. Includes a fix for H3 speaker/Shot-label compatibility, plus support for MOSS Image, Flux 3 LoRAs, Viggle, face-refiner repairs, and Recast memory improvements.
@electra_highwayCannot select LoRAs for Flux 3 Klein. The Advanced Settings section shows the option labeled "Presets" instead of "LoRAs & presets" as with other models.
@blizaine·1mo ago
in Maestro
Just pushed v3.1 to the Public Dev branch for testing. Push to Main should be coming later today if all goes well. Includes a fix for H3 speaker/Shot-label compatibility, plus support for MOSS Image, Flux 3 LoRAs, Viggle, face-refiner repairs, and Recast memory improvements.
@treedeaViggle Animate error because "it not exist" and using Recast is out of possibilities - with 4080 and 32 RAM 5 seconds clip taking like 20 hours to process
@blizaine·1mo ago
in Maestro
Just pushed v3.1 to the Public Dev branch for testing. Push to Main should be coming later today if all goes well. Includes a fix for H3 speaker/Shot-label compatibility, plus support for MOSS Image, Flux 3 LoRAs, Viggle, face-refiner repairs, and Recast memory improvements.
@nambusI've been having trouble with Maestro. It’s taking hours just for a 13.7 second video clip using the studio tab even after I have updated to 2.6.0 version. I’m using the H3 Omni Prunned model. Prior to updating with the 2.4.2 version a 5 second video would take 5 minutes but a 10 second video took an hour and 46 minut
@blizaine·1mo ago
in Maestro
I've got a nice memory optimization update coming very soon. final testing is happening today. I was able to acquire a second test machine with less RAM and VRAM than my primary machine and I've been having my agent run non stop optimization tests on it for the last 3 days or so.
@progHis report is accurate. I have the same specs and what once took 5 hours for a 5 min video is now taking 19 hours. Same prompts by and same models. I’m on i9 rtx Nvidia 5070ti 12gb with 32 ram. Windows 11.
@blizaine·1mo ago
in Maestro
was your 5hr example of Director mode generation? approx duration of video? I just set up a second test machine that I'm having my agent run tests on and optimize Maestro for and it is a 12GB 3080ti with 32GB of Ram.
@guardian123i dont get it, i can use director mode and if its 30 seconds or 1 min to generate a movie, its fine. Problem is that when i try 10 mins or more, the director mode is stuck on "calibrating ETA" and doesnt move or fail or anything. I checked the terminal and it saids GPU memory poor 3.8.2. Im not sure is it because
@blizaine·1mo ago
in Maestro
it should work. must be a bug. how much RAM and VRAM do you have? I have a second test machine I got set up for testing yesterday with less VRAM & RAM than my main, so I'm working on some fixes and optimizations.
@cynic2020All good. For my liking, it’s too bloated, as I already stated, and changing the settings is also a hassle if you just want to quickly spin up a short video. All that automatic stuff—like the Qwen model, which I wasn’t going to use anyway—also gets shoved into the automatic loader the first time it starts up. As a Lin
@blizaine·1mo ago
in Maestro
Good feedback. I may set up my other machine as a dual-boot Win/Linux.
@cynic2020Bloated and unusable. It used to function properly when Mini came out—that was when I first installed it. Back then, it took around 10–12 minutes to generate something; now, it gives me estimated times of about five hours. It also comes with a lot of unnecessary content preloaded. Thanks. I’ve already removed it, so
@blizaine·1mo ago
in Maestro
Sorry to hear that. I just got a test machine with less VRAM, so I can optimize Maestro better. Also, I'm porting over Wan2GP 17's new optimized memory manager.
You’re all caught up.
Pinokio
PrivacyTerms
FeedAppsCreateRemote