A new version of the popular Deforum Stable Diffusion notebook has just been updated to version 0.5 with some very cool and new additions!
- Display of Animated Values
- Option to Overwrite Extracted Frames in Video Input Mode
- Perspective 2D Flipping
- Import Settings File
- Custom MATH expressions
- Output Steps
- Dynamic Video Masking
- Weighted Prompts
- Auto Download of Models
- Waifu and Robo Diffusion Models
Here is a breakdown from the Google Doc as to what each new feature does!
We have also added those changes to our guide located here.
Display of Animated Values:
During a render, information will be displayed alongside the progress bar that shows current Angle, Zoom, Translation and Rotation values used to calculate the movement for the current frame.
Option to Overwrite Extracted Frames in Video Input Mode:
By default, a video file will extract and save its frames to drive every run. This new option allows the user to bypass this process for future runs, and skip right ahead to render.
Perspective 2D Flipping:
This feature allows extra parameters during 2D mode to allow a “faux” Roll, Tilt, Pan canvas function only found in 3D mode. Users may use “theta, phi & gamma” angle control to simulate a 2.5D effect, using only a 2D canvas mode.
Import Settings File:
Users may now select an “override” function that will bypass all instructions from the notebook settings, and instead run from a settings.txt file previously saved by the user. This function is reverse compatible to v04. This feature does not auto-populate settings into your notebook, however it directly runs the instructions found within the .txt file.
Custom MATH expressions:
Users may now use custom math expressions as well as typical values as scheduling for parameters that allow strings, such as zoom, angle, translation, rotation, strength_schdule, and noise” Many wave functions can now be achieved with simple instructions using “t” as a variable to express frame number. Please refer to the link provided for more info about math functions.
Users may now choose to view intermediate steps of a frame, as well as the option to save those steps as output images to drive. This powerful feature may use a lot of drive space, and browser cache if many steps are used in long renders. 500 steps will display/save 500 images.
Dynamic Video Masking:
During Video Input mode, users may select to also include an additional video to be used as a mask. Frames will be extracted for both the video init, as well as the video mask, and used in conjunction.
A numerical prompt weight feature has been added to deforum as a selectable feature. When enabled, the run will interpret the values and weights syntax of the prompt for better control and token presence. Bonus: weights values adhere to MATH expressions for even more control.
“this prompt has weights if prompt weighting enabled:2 can also do negative:-2”, # (see prompt_weighting in the notebook)
Auto Download of Models:
Users that have successfully created a huggingface account, and accepted license/usage agreements will be able to skip manual download of .ckpt model files. Bonus: We’ve included “Waifu” and “Robo” fine-tuned models as well, to select from.
“a nousr robot, trending on Artstation”, # use “nousr robot” with the robot diffusion model (see model_checkpoint setting in notebook dropdown)
“touhou 1girl komeiji_koishi portrait, green hair”, # waifu diffusion prompts can use danbooru tag groups (see model_checkpoint in notebook dropdown)