Turn your picture into a Lego character(youml.com)
youml.com
Turn your picture into a Lego character
https://youml.com/recipes/236-photo-to-lego-minifigure
16 comments
Seems to be hugged to death with an OpenAI rate limit.
You guys broke it! :)
Apparently there is a 7-image per minute hard limit on OpenAI image generation, so come back a bit later.
Apparently there is a 7-image per minute hard limit on OpenAI image generation, so come back a bit later.
That's odd IMO since none of this workflow should be hitting OpenAI.
Error code: 429 - {'error': {'message': 'Rate limit reached for gpt-4-vision-preview in organization org-eRnbbdbvRysampoMyGNkYZ8A on tokens per min (TPM): Limit 10000, Used 6380, Requested 4151. Please try again in 3.186s. Visit https://platform.openai.com/account/rate-limits to learn more.', 'type': 'tokens', 'param': None, 'code': 'rate_limit_exceeded'}}
They're using GPT-4-Vision? With its 100-requests-a-day rate limit? Huh.
Why restrict it to only Google accounts?
It's a cheap way to prevent abuse.
If you have no log in flow at all you'll be spammed. If you have your own sign in flow how do you detect and stop automated account creation and subsequent spam? Can you beat Google at stopping bot signups?
In the end it's a lightweight barrier to abuse. You can argue they should diversify and not just use Google but hey it's a small project. It's understandable. It's also free to sign up for Google for a real user.
If you have no log in flow at all you'll be spammed. If you have your own sign in flow how do you detect and stop automated account creation and subsequent spam? Can you beat Google at stopping bot signups?
In the end it's a lightweight barrier to abuse. You can argue they should diversify and not just use Google but hey it's a small project. It's understandable. It's also free to sign up for Google for a real user.
Ever so slightly harder to create dozens of accounts to get free credits?
Presumably to harvest sign-in information
didn't work for me :/ got a safety system error
yeah... 429 rate limited requests... and dennnnnnn out of credits :(
This one right here, officer /s
Some notes on how you want to do this yourself:
1. Get atleast ten 1024x1024 images of a style you want (e.g. Lego). You can use SDXL/Midjourney to generate these if you are good at curation.
2. Train a LoRA on SDXL using those images.
3. Construct an image-to-image pipeline with the LoRA'd SDXL and pass along the input image as the base image. You can run for fewer steps (20-30 instead of 50) and/or preprocess the input image (e.g. segmentation).
There is a lot of other parameter tweaking for both training and generation which is where the added value is for websites such as these. But with a service like Replicate the pipeline is relatively cheap (~$0.50 for training, $0.01 for each image generation). And will only get much cheaper once people play around with LoRAs for smaller and faster models like ssd-1m-lcm.
1. Get atleast ten 1024x1024 images of a style you want (e.g. Lego). You can use SDXL/Midjourney to generate these if you are good at curation.
2. Train a LoRA on SDXL using those images.
3. Construct an image-to-image pipeline with the LoRA'd SDXL and pass along the input image as the base image. You can run for fewer steps (20-30 instead of 50) and/or preprocess the input image (e.g. segmentation).
There is a lot of other parameter tweaking for both training and generation which is where the added value is for websites such as these. But with a service like Replicate the pipeline is relatively cheap (~$0.50 for training, $0.01 for each image generation). And will only get much cheaper once people play around with LoRAs for smaller and faster models like ssd-1m-lcm.
Good thing it's free, because it failed but still took my credits.