Updates to existing tools

Prompt builder

The Prompt Builder tool has been updated. In addition to an updated design, you can now add a prefix and suffix to both the prompt and negative prompt, which is great for adding embedding and lora triggers to your generated prompts.

In the previous version you could randomize most variables - this functionality has been extended to allow for random categories, genres, colors and styles.

One nice use for this tool is to set the seed to random, choose random variables, and then generate a large sample size and automatically generate a large set of completely random images that you can sift through afterwards.

Prompt browser

The prompt browser is back. You can save prompts from the File > Save menu, then view your saved prompts by clicking Tools > Prompt Browser which will display a popup window containing your saved prompts. You can then load them into your current tab, delete them, or now (with AI Runner v2.2.0) edit them. Changes you make to your prompt collection are automatically saved.

Spinboxes are back

Spinboxes for all sliders were removed in AI Runner v2.0.0, however our users like them, and so do we - so they’re back.

New Features

Text-to-video with Zeroscope and Zero Shot

AI Runner v2.1.0 introduces support for text-2-video (txt2vid) using Zeroscope V2 and Zeroscope V2 XL models in addition to Zero Shot. This is a work in progress and will improve over time, however it is currently considered an experimental feature. Although you can generate short videos using 8gb CUDA capable GPUs, anything over a 2 second clip causes memory errors, so it is recommended that you have a video card with much more VRAM.

Shap-e gif generation

Support for the openai Shar-E 3d gif model has been added and can be found to the right of the Kandinsky tab.

Use Shap-e to generate 3d animated gifs from text prompts, or from existing images… but don’t expect great results. The base models generate fairly low quality images, (and they take quite a while to produce on 8gb cards).

Bug fixes and minor additional features

AI Runner 2.2.0 patches a large number of bugs and has some minor feature updates as well.

Additional features

• MUCH faster CKPT / Safetensor loading
• When switching between txt2img / img2img and using the same model, the model is no longer reloaded from scratch
• Custom txt2vid path support added to path settings
• Custom gif path support added to path settings
• Adds clothing to prompt variables
• Removed some NSFW variables that were generated with an LLM and mistakenly added to prompt builder
• Adds clothing and a more comprehensive list of ethnicity variables to the person category of prompt builder
• Adds more food options to prompt builder variables
• Adds a new action to the toolbar which allows you to clear all prompts from all tabs
• Embeddings that fail to load for a given model will now display as deactivated for that specific model
• The program will no longer hang if running AI Runner from command prompt / terminal and pressing ctrl+c to early exit
• Removes bitsandbytes and triton from required libraries

Bug fixes

• Active grid border color added for txt2vid
• Fixes a bug which would cause an error when attempting to add embeddings to prompts by clicking the embedding buttons
• Removes metadata import from images which was preventing import of images previously exported with metadata
• File new, open and save bugs have been fixed
• Layer deletion bugs have been fixed
• Progress bar bugs have been fixed
• Prompt builder toggling has been fixed
• A bug which prevent all elements from updating when toggle light / darkmode has been fixed