Locopic
Locopic started as an experiment with AI tools. Generative AI needs human-created data to get going, but once it does can it create accidentally brilliant images? Can AI create viral images on its own?Let's have a go. Check out the first batch of images below that were fully generated by a connected workflow of AI tools.
A series of AI tools are stacked up in a workflow:Step 1. ChatGPT is instructed to list options for 3 sets of variables: a subject, a stylistic effect, and a format.
Step 2. These variables are input into the OpenAI Playground where the GPT is asked to select a combination that will generate an interesting image.
Step 3. The GPT generates a prompt that is generated in Midjourney, which is then upscaled and posted to Twitter.//Update October 2023:- Increased the amount of variables and art styles.
- Added film stocks.
- Used Playground AI to create an agent to select what it thinks will be most interesting and visually appealing combination of variables.
- Noticed that too many of each variable (200+) means that it's easier to go too random and land on a chaotic image like the baseball tornado below.
- The simpler the prompt the better the output.
//Update December 2023:Since MidJourney v6 was released, it was worth revisiting and trying new outputs to see what this new version could do.- Used ChatGPT to add more variables and art styles.
- new prompt format: Subject | Setting | Style | Composition | Lighting | Film Stocks
- Surprisingly the outputs were not as strong when lots of details are added to the Midjourney prompts. Too much emphasis on lighting, composition, and setting, would create fussy images.
- Midjourney v6 is very good at details and lighting.
- The upscaling in Midjourney v6 works extremely well!Results below.
This has been just the start. Follow Locopic on X (formerly Twitter) for more images, videos, and collaborations.