Important excerpt about censorship from Ideograms technical blog
> The reference pipeline validates every prompt against the JSON schema before generation and rejects inputs that do not parse, so the input format at inference time is the same one the model saw during training.
It seems they simply used the same safety message for validating the prompt is something the model is trained on (valid JSON), so when the model is rejecting simple prompts it's not due to wanting to censor something like "cat" but to make sure the prompt is valid.
Not sure if this is a good approach to enforcing "good" prompting for a model but it makes a lot more sense than some crazy levels of censorship and false positives.
https://ideogram.ai/blog/ideogram-4.0/
https://redd.it/1twychx
@rStableDiffusion
> The reference pipeline validates every prompt against the JSON schema before generation and rejects inputs that do not parse, so the input format at inference time is the same one the model saw during training.
It seems they simply used the same safety message for validating the prompt is something the model is trained on (valid JSON), so when the model is rejecting simple prompts it's not due to wanting to censor something like "cat" but to make sure the prompt is valid.
Not sure if this is a good approach to enforcing "good" prompting for a model but it makes a lot more sense than some crazy levels of censorship and false positives.
https://ideogram.ai/blog/ideogram-4.0/
https://redd.it/1twychx
@rStableDiffusion
Ideogram
Ideogram 4.0 Technical Details: Open model at the forefront of design
Our first open-weight foundation model. A 9.3B single-stream Diffusion Transformer, trained from scratch, with a vision-language text encoder and structured JSON prompts.
Announcing Comfy Desktop: One App for every Comfy, rolling out 100% by Monday June 8
https://redd.it/1tx4wsm
@rStableDiffusion
https://redd.it/1tx4wsm
@rStableDiffusion
This media is not supported in your browser
VIEW IN TELEGRAM
hildegard - tiled upscaling and refining based on flux 2 klein
https://redd.it/1tx59t3
@rStableDiffusion
https://redd.it/1tx59t3
@rStableDiffusion
ComfyUI-PiD update: more backbones, workflows, and better low-VRAM support
https://redd.it/1txacn4
@rStableDiffusion
https://redd.it/1txacn4
@rStableDiffusion
Ideogram generated a Gemini Watermark without being prompted to
https://redd.it/1txfrhw
@rStableDiffusion
https://redd.it/1txfrhw
@rStableDiffusion
I didn't expect ideogram to be so good
https://preview.redd.it/odhzj8racf5h1.png?width=1501&format=png&auto=webp&s=71dbe0a0d613fb00dc8a904cc58646b7639bf02b
Spanish speaker trying to make their first contribution, please excuse any poor writing.
With Ideogram's release on Comfy, I saw it received a lot of hate, but honestly, it's amazing (at least to me) what it can do regarding typography. I adapted a workflow that was shared in the comments and combined it with Flux image editing, and wow, the possibilities are enormous.
First, I was interested in optimizing workflows for the hypothetical creation of content to promote an e-commerce site or product. I don't know if this violates the model's usage guidelines, but it works perfectly.
Second... yes... the workflow can be other things... , and it's just a matter of using the JSON prompt correctly. Otherwise, with Flux, you can do character replacement, and it looks quite nice.
I left the workflow with 30 steps and 5 CFGs on the Ideogram side, and it works wonderfully for typography and other details (wink wink). I don't know if you've tried other values, but with these and a resolution of 1024 (I'll upscale it later), the total generation time between both models (Ideogram and Flux) is 125 seconds. My setup is a 5070 and 96GB of RAM, and considering that it basically renders the product already finished, I find it truly impressive for the time saved when adding details in an editing program.
Here's a comparison of how this layout process was before in Flux (40 steps and 4 configurations, almost 15 minutes of generation) and how it is now in Ideogram.
Image generated by Infogram
image generated by flux
image with the flux edit
product image
I've included the workflow here so you can save yourself the trouble of switching between different workflows and see what you can create. Again, I simply combined two existing workflows to save time until someone achieves an i2i.
https://pastebin.com/r6UvLyni
Note: It seems I messed up a node; just activate it for the Infogram side to work.
https://preview.redd.it/hrle6nx3if5h1.png?width=565&format=png&auto=webp&s=a60e90a11716b5fccb714909b47bae9dec369fde
https://redd.it/1txf7qe
@rStableDiffusion
https://preview.redd.it/odhzj8racf5h1.png?width=1501&format=png&auto=webp&s=71dbe0a0d613fb00dc8a904cc58646b7639bf02b
Spanish speaker trying to make their first contribution, please excuse any poor writing.
With Ideogram's release on Comfy, I saw it received a lot of hate, but honestly, it's amazing (at least to me) what it can do regarding typography. I adapted a workflow that was shared in the comments and combined it with Flux image editing, and wow, the possibilities are enormous.
First, I was interested in optimizing workflows for the hypothetical creation of content to promote an e-commerce site or product. I don't know if this violates the model's usage guidelines, but it works perfectly.
Second... yes... the workflow can be other things... , and it's just a matter of using the JSON prompt correctly. Otherwise, with Flux, you can do character replacement, and it looks quite nice.
I left the workflow with 30 steps and 5 CFGs on the Ideogram side, and it works wonderfully for typography and other details (wink wink). I don't know if you've tried other values, but with these and a resolution of 1024 (I'll upscale it later), the total generation time between both models (Ideogram and Flux) is 125 seconds. My setup is a 5070 and 96GB of RAM, and considering that it basically renders the product already finished, I find it truly impressive for the time saved when adding details in an editing program.
Here's a comparison of how this layout process was before in Flux (40 steps and 4 configurations, almost 15 minutes of generation) and how it is now in Ideogram.
Image generated by Infogram
image generated by flux
image with the flux edit
product image
I've included the workflow here so you can save yourself the trouble of switching between different workflows and see what you can create. Again, I simply combined two existing workflows to save time until someone achieves an i2i.
https://pastebin.com/r6UvLyni
Note: It seems I messed up a node; just activate it for the Infogram side to work.
https://preview.redd.it/hrle6nx3if5h1.png?width=565&format=png&auto=webp&s=a60e90a11716b5fccb714909b47bae9dec369fde
https://redd.it/1txf7qe
@rStableDiffusion
Lightricks to split into two companies as it cuts another 75 jobs
https://www.calcalistech.com/ctechnews/article/r1dgjt5gmg
https://redd.it/1txhfv7
@rStableDiffusion
https://www.calcalistech.com/ctechnews/article/r1dgjt5gmg
https://redd.it/1txhfv7
@rStableDiffusion
ctech
Lightricks to split into two companies as it cuts another 75 jobs
Jerusalem unicorn reshapes itself around AI, as Facetune profitability funds a costly push into video models.