5 ms·

From the cherry-picked example-images on that page, it seems like Imagen more closely follows the prompt than the open Stable Diffusion model[0]. Stable Diffusi

by tomthe 4y ago

From the cherry-picked example-images on that page, it seems like Imagen more closely follows the prompt than the open Stable Diffusion model[0]. Stable Diffusion needs a lot of hints before it makes out of the ordinary pictures. In general, I think these models are a great and funny toy, but not a threat to stock-photos yet. This may change within a year or three years though. [0]:https://stability.ai/blog/stable-diffusion-announcement https://stability.ai/blog/stable-diffusion-announcement

culi 4y ago
Not a threat to stock photos? That's exactly what they are. Look at these photos from the Midjourney Discord today: crystal dragon thing: https://cdn.discordapp.com/attachments/951197655021797436/1009511356342489138/1.png https://cdn.discordapp.com/attachments/951197655021797436/10... https://cdn.discordapp.com/attachments/951197655021797436/1009511359131693056/14.png https://cdn.discordapp.com/attachments/951197655021797436/10... https://cdn.discordapp.com/attachments/951197655021797436/1009511358745804881/13.png https://cdn.discordapp.com/attachments/951197655021797436/10... davinci-style notebook of flying machines: https://cdn.discordapp.com/attachments/1008049109338443829/1011122626833420298/20220708_013655.jpg https://cdn.discordapp.com/attachments/1008049109338443829/1... https://cdn.discordapp.com/attachments/1008049109338443829/1011122627772940398/20220708_025002.jpg https://cdn.discordapp.com/attachments/1008049109338443829/1... this person tried to show the life cycle of an alien: https://cdn.discordapp.com/attachments/1010211132671275058/1010954794912981002/01.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954795298852995/02.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954795961561138/04.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954796657807410/06.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... cavemen taking a group selfie (lots of faces) https://cdn.discordapp.com/attachments/1011408429170044928/1011620969544159393/fifty_35mm_close-up_photo_of_faces_of_cavemen_taking_a_group_se_63e714d1-b5a8-43d3-9025-486bb6f0f8be.png https://cdn.discordapp.com/attachments/1011408429170044928/1...
- sh4rks 4y ago
  The davinci ones are incredibly intricate
- 8n4vidtmkvmk 4y ago
  those were generated? those look dope
  - culi 4y ago
    Generated and paying for MidJourney gives you the copyright to them so you can use them for whatever projects you want
    - glenneroo 4y ago
      Which feels a bit sketchy(?) to me, seeing as all models are built on imagery scraped from the net without anyone's permission. It's one reason why I've spent the last month training my own models using my own imagery. If these stock photo sites had any brains, they would also start training models on images in their databases, especially since they already have everything sorted into categories based on keywords (which I'll spend the next year doing, until I can get img2text tools working in recursive batch mode).
      - culi 4y ago
        I totally agree. Also, that sounds like a wonderful project :) You should post project updates somewhere to keep track of your model as it evolves!
        glenneroo 4y ago
        Thanks! I have a g-doc where I've been documenting settings and progress here[0] although I just now realized I might not have been using my diffusion model correctly for most of my tests. Iteration 509 of my model I seem to have finally nailed it though! :) I partially "blame" Visions of Chaos since the amazing dev (or devs?) drops updates almost every day with new Machine Learning features, model training was only added recently. I must have reset something on accident. Also I realize there's a lot of image prep work required, not to mention I have a less than ideal amount of VRAM (3060 Ti w/8GB but no monitors attached i.e. 8GB free) so I have to lower some settings. The source images have to be in 1:1 format (which none of my photos are) so I'm using a script to batch call ImageMagick's 'convert' to add white borders to the top/bottom, which results in my renders also having white borders. [0] https://docs.google.com/document/d/1CnC5SaqpeJiQS-TlDS4trzJRxTfnETdASWhi4I3xu9I/edit?usp=sharing https://docs.google.com/document/d/1CnC5SaqpeJiQS-TlDS4trzJR...

by tomthe 4y ago

culi 4y ago

Not a threat to stock photos? That's exactly what they are. Look at these photos from the Midjourney Discord today: crystal dragon thing: https://cdn.discordapp.com/attachments/951197655021797436/1009511356342489138/1.png https://cdn.discordapp.com/attachments/951197655021797436/10... https://cdn.discordapp.com/attachments/951197655021797436/1009511359131693056/14.png https://cdn.discordapp.com/attachments/951197655021797436/10... https://cdn.discordapp.com/attachments/951197655021797436/1009511358745804881/13.png https://cdn.discordapp.com/attachments/951197655021797436/10... davinci-style notebook of flying machines: https://cdn.discordapp.com/attachments/1008049109338443829/1011122626833420298/20220708_013655.jpg https://cdn.discordapp.com/attachments/1008049109338443829/1... https://cdn.discordapp.com/attachments/1008049109338443829/1011122627772940398/20220708_025002.jpg https://cdn.discordapp.com/attachments/1008049109338443829/1... this person tried to show the life cycle of an alien: https://cdn.discordapp.com/attachments/1010211132671275058/1010954794912981002/01.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954795298852995/02.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954795961561138/04.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... https://cdn.discordapp.com/attachments/1010211132671275058/1010954796657807410/06.jpg https://cdn.discordapp.com/attachments/1010211132671275058/1... cavemen taking a group selfie (lots of faces) https://cdn.discordapp.com/attachments/1011408429170044928/1011620969544159393/fifty_35mm_close-up_photo_of_faces_of_cavemen_taking_a_group_se_63e714d1-b5a8-43d3-9025-486bb6f0f8be.png https://cdn.discordapp.com/attachments/1011408429170044928/1...

sh4rks 4y ago
The davinci ones are incredibly intricate
8n4vidtmkvmk 4y ago
those were generated? those look dope
- culi 4y ago
  Generated and paying for MidJourney gives you the copyright to them so you can use them for whatever projects you want
  - glenneroo 4y ago
    Which feels a bit sketchy(?) to me, seeing as all models are built on imagery scraped from the net without anyone's permission. It's one reason why I've spent the last month training my own models using my own imagery. If these stock photo sites had any brains, they would also start training models on images in their databases, especially since they already have everything sorted into categories based on keywords (which I'll spend the next year doing, until I can get img2text tools working in recursive batch mode).
    - culi 4y ago
      I totally agree. Also, that sounds like a wonderful project :) You should post project updates somewhere to keep track of your model as it evolves!
      - glenneroo 4y ago
        Thanks! I have a g-doc where I've been documenting settings and progress here[0] although I just now realized I might not have been using my diffusion model correctly for most of my tests. Iteration 509 of my model I seem to have finally nailed it though! :) I partially "blame" Visions of Chaos since the amazing dev (or devs?) drops updates almost every day with new Machine Learning features, model training was only added recently. I must have reset something on accident. Also I realize there's a lot of image prep work required, not to mention I have a less than ideal amount of VRAM (3060 Ti w/8GB but no monitors attached i.e. 8GB free) so I have to lower some settings. The source images have to be in 1:1 format (which none of my photos are) so I'm using a script to batch call ImageMagick's 'convert' to add white borders to the top/bottom, which results in my renders also having white borders. [0] https://docs.google.com/document/d/1CnC5SaqpeJiQS-TlDS4trzJRxTfnETdASWhi4I3xu9I/edit?usp=sharing https://docs.google.com/document/d/1CnC5SaqpeJiQS-TlDS4trzJR...