Community Update
I hope you like the new design and find things easier to use. Graydient is growing and we’re really lucky and thankful for your support. We mean that. Most of the software on Graydient was created by our founder Tlack305, we’re as indie as they come. This space was intended to tell you about our software updates but just wanted to take a minute to thank to everyone who has stuck with us in spite of Patreon deleting our page. We got kicked down, but we got up again, thanks to you!
Software Update
Stable Diffusion 3.5 Large and Turbo are now available
Use them in Unified or Telegram, check our Workflows.
Ron Rider | more featured works
Roadmap Update
Current focus: In order of priority: (1) Dynamic Flux Lora loading (2) Adding a toggle for dark mode and (3) the ability to filter and hide concepts (4) Quality of Life UI/UX all around like upgrading Inpaint to Flux and making Img2Img easier to use. Also coming up is a new customizable dashboard called My Console. It’s a rough wip, give us a little time and it will become beautious & useful.
Help us prioritize the features that you want first by Posting Suggestions and voting up the best ideas. By the way, are missing a model/concept? You can Request an install. Self-service model installs and videos/animation is also a stretch goal for us, but the fastest Gen5 NVMEes don’t grow on trees, so please tell a friend about Graydient 🙂
Chat with me: Our dev team is online everyday on Telegram sharing prompt ideas, taking model requests, and answering support questions. Join our VIP chat group and open the Topics panel to see threads on prompt engineering, news, announcements, videos, and more.
Learn more about what we’re building: check the Graydient Blog and our Telegram Announcements Channel
Baddie’s Corner: Community Questions
Question: How do I make a consistent character?
Method #1: The quick and dirty
Give them a name and nationality and age at the start of the prompt, then keep your prompt parameters the same: Seed, Guidance, Steps, and Concepts/Models. Put the most important parts of the character’s look at the very front of the prompt, and never move it. Then change the tail end of the prompt and inch the seed value by 1 to get variations. This techique is called “seed travel” and requires very little effort or prep time.
Method #2: The technical way
The intermediate method is to use a fixed set of concepts (LoRas, Embeddings, Checkpoints) to lock in the look of character, or train your own LoRa. You can mix and match loras at lower weights to make a unique persona. Combine this technique with the quick and dirty method for best results. This takes some practice, but it’s how the pros are doing believable characters.
To get two consistent characters in a scene like this, describe the other character in the initial prompt, and then use our Inpaint tool to “paint” a mask over the off-brand character with your technical prompt for character #2. You’re adding them in the inpaint step to achieve consistency. We recommend learning how to use our Controlnet tool to nail their poses without complicated prompts. You can grab a pose off the internet and use it like a skeleton/stencil, for one character or both.
As this activity is becoming increasingly popular, our dev team is working on a more convenient workflow to do this faster, like regional prompting and bringing facelock/facepush to our Unified editor.
Become a Genius
Fancy a personal 1-on-1 AI coach? Talk to a human over video or Telegram chat for a full hour, and request a class on any curiosity. It’s an optional, additional service that you can book today – Browse AI coaches. Space is limited, so please check our available time slots. What can our coaches do? They can train a Flux Lora for you, build a custom AI chat bot for you, and much more. We’re all adults here, you can ask about stuff. Of course, we also offer many free resources: Get prompt tips on our Subreddit and Youtube, and a separate youtube for Telegram users called the PirateDiffusion Show
That’s all for this blog!
