POM
@peteromallet
Followers
6K
Following
14K
Media
2K
Statuses
9K
Working with a ragtag group of nerds to accelerate The 2nd Renaissance, @banodoco (https://t.co/K3EvyToS2N). π¬: @hannahsubmarine
Berlin/Ireland
Joined December 2011
Sharing InStyle, a LoRA for Qwen-Edit that allows it to generate images based on a style reference. The base model can attempt this task but has frequent issues. This LoRA helps it properly understand styles + avoid non-style references to the input. Link + dataset below!
7
34
236
.@ChetiArt quietly hacking together different models and ideas from the open source community to create the greatest video upscaler in the world:
Me: Can I upscale these GTA5 gifs so they look real? Wan2.2: Hold my beer Made with @Alibaba_Wan + @ComfyUI and Cseti's killer upscaling workflow π
0
0
10
First they politely ignore you, then they passively help you, then they actively support you, then you both win.
1
0
3
I just did a bunch of ugly hacks trusting that LLMs will get good enough that I can fix them easily before they become a problem
1
0
2
we should be way more excited about Ovi π₯ > open source, apache 2.0 license > text/image-to-video + audio generation > based on Wan2.2 5B > great latency <> quality balance
7
20
135
The best thing about making an AI art tool is when your output suddenly gets more beautiful when you fix a subtle bug or misconfigured param
I've been building a new tool and attached are two of the first creations out of it Both are crappy, but the one on the right is less crappy than the previous one. This is how you know that using it is a skill
0
0
7
A single division of a single Chinese company is defeating the West
1
0
2
This is for this year's version of my annual 'Sound of the Tires in the Snow' video, This will be my four year making it - my goal is to do an annual checkpoint of both AI + my own skills. The first year I made it, 1 person messaged me saying that it had a big impact on them,
I've been building a new tool and attached are two of the first creations out of it Both are crappy, but the one on the right is less crappy than the previous one. This is how you know that using it is a skill
0
0
3
This is a very complex shot that I have a very precise idea of and I know that I'll be able to make it exactly as I imagine!
0
0
4
I've been building a new tool and attached are two of the first creations out of it Both are crappy, but the one on the right is less crappy than the previous one. This is how you know that using it is a skill
1
0
29
A few months ago, I had an idea for a film and started working on a trailer in my spare time. Excited to finally share it's complete! π¬ Some shots are four months old, but I evolved alongside the latest video gen models, using @Kling_ai extensively. The new 2.5 Turbo model
32
22
191
the π³ππΉπ¨ π£π―π’π·π«π¦ is LIVE Train Qwen, Wan and FLUX LoRAs for free for 1 week (Sep 29 - Oct 6th) We cobbled together @ostrisai AI Toolkit & the new @huggingface Jobs API together
25
23
143
I built a Sora 2-inspired video pipeline in ComfyUI and you can download it ! Technical approach: β 4 LLMs pre-process everything (dialogue, shot composition, animation direction, voice profile) β Scene 1: Generate image with Qwen-Image β automated face swap (reference
12
41
347
I would prefer to be charitable but I just canβt imagine a scenario where they somehow mistook Wan 2.5 for Sora 2 from dev -> prod -> marketing Absolutely soulless company
1
1
4
Did you know that you can get a month of free Higgsfield credits by retweeting a post where they lie about having Sora 2?
so @higgsfield_ai advertised having access to Sora 2. Reached out to content creators to PROMOTE Sora being on their. But really it was Wan 2.5 with Sora name slapped on it! That can't be legal lmao. How greasy! Care to comment @higgsfield_ai or was it an 'honest mistake'.
1
0
7
For everything thatβs lovely is But a brief, dreamy, kind delight. O always give the heart outright
0
0
3