JAFFA's Custom Deforum Workflow MAY 2026 DPM++3M SDE Expo

This is a workflow that utilises Deforum Nodes in ComfyUI to create various images that convey a somewhat linear animation for the use of videos. I currently use this workflow to create backgrounds on my music channel on YouTube I have a discord if you would like to join to discuss AI, the workflow, music, play games or whatever else you can join here; https://discord.gg/U5Ep2BTsfr Out of the box this model generates SFW images. Negative prompts include some NSFW terms to keep images SFW. But of course in your own hands you can change whatever you like and generate whatever you want with whatever checkpoints you want. Hurray, freedom! If you want to support myself or this model all I ask is that you throw a free subscription to my Unmonetized Music channel on YouTube. ♥ This workflow can be used in conjunction with the following tools which you may find useful; https://civitai.com/articles/14641 https://civitai.com/models/613826/sdxl-imagepath2vid-output-upscale-workflow Changelog: VERSION 15.0: Changed Scheduler to DPM++3M SDE Expo for better quality/coherence. VERSION 14.0: ComfyUI Desktop Migration Version. VERSION 13.0: Small August Update, small changes, nothing important, just an update. VERSION 12.0: Added a Custom Strength Schedule as explained here; https://civitai.com/articles/14641 VERSION 11.0: Few Variation additions with small notes for guidance. Expanded Interpolation to 6 total frames instead of 4. Changed Hybrid Motion to Farneback for quality gain in interpolation VERSION 10.0: DPM++ 2M Karras ZavyChroma v.10 Optimised for personal use. (TRANSLATE IS BROKEN on ALL versions. IF ANYONE KNOWS HOW TO FIX PLEASE SEND ME A MESSAGE) VERSION 8.0: SDXL 1.0 Deforum Video Creation Workflow. DDPM VERSION 5.0 Onwards: SDXL 1.0 Deforum Video Creation Workflow. DPM++ SDE Karras Perfect for use with ZavyChromaXL Checkpoint. Strength is low due to reduce flickering. Optimum CFG is 5-7 High detail, high depth. Added -2 Clip Skip Added VAE loader Added Upscale Model Loader Added Lora Capability Refined Upscaler to use Ultimate SD Upscaler instead of default lanczos. Upscaler for 1920x1080 resolution (Can be bumped up to around 4k) Previous Versions have notes and the version number will have direct changes beside it) (Details may diminish over time with low Strength, recommend bumping to: 0.25-0.49 if you see detail disappear.) (When starting generation disable the Upscale node, Preview Image node and Save Image and Save Video node until one frame has been generated for the cache to pull from, after first generation enable all above nodes and all should be good!) If you want to save frames, use Save Image Extended, if you want to create video use Save Video. Working on this for the past 3 months. Finally got to a place I'm happy. Flickering is still present as Deforum is temperamental, until I find a different selection of nodes to achieve the same thing this is probably the best I can achieve with Deforum. But who knows, there may be future iterations of this workflow. Thanks for all the support on this journey and feel free to use and post anything you create with this workflow alongside the Gallery of this submission. Appreciate all the buzz and support and downloads. Thankyou! ♥ This Deforum ComfyUI workflow is my test for generating backgrounds for YouTube videos over here .