Skip to content

Latest commit

 

History

History
506 lines (424 loc) · 41.4 KB

Video Model Support.md

File metadata and controls

506 lines (424 loc) · 41.4 KB

Video Model Type Support In SwarmUI

Model Year Author Scale Type Censored? Quality/Status
Hunyuan Video 2024 Tencent 12B MMDiT Text2Video and Image2Video variants No Modern, Decent Quality
Hunyuan Video 1.5 2025 Tencent 8B MMDiT Text2Video and Image2Video variants No Modern, Decent Quality
Lightricks LTX Video 2024 Lightricks 3B DiT Text/Image 2Video ? Modern, Fast but ugly
Lightricks LTX Video 2 2026 Lightricks 19B DiT Text/Image 2Video+Audio Minimal Modern, good/mixed quality but fun
Wan 2.1 and 2.2 2025 Alibaba - Wan-AI 1.3B, 5B, 14B Text/Image 2Video No Modern, Incredible Quality
Kandinsky 5 2025 Kandinsky Lab 2B, 19B Text/Image 2Video No Modern, Decent Quality

Support for image models and technical formats is documented in the Model Support doc, as well as explanation of the table columns above

Old or bad options also tracked listed:

Model Year Author Scale Type Censored? Quality/Status
Stable Video Diffusion 2023 Stability AI 1B Unet Image2Video Yes Outdated
Genmo Mochi 1 2024 Genmo 10B DiT Text2Video ? Outdated
Nvidia Cosmos 2025 NVIDIA Various Text/Image/Video 2Video ? Modern, very slow, poor quality

Current Recommendations

Video model(s) most worth using, as of December 2025:

  • Wan 2.2 or 2.1, in 14B either way. It's the best you can get locally currently.
  • Kandinsky 19B looks interesting, but is new and struggling to reach its potential. Could be worth playing with.

Demo Gifs

  • Video demos included below are seed 1 of the prompt wide shot, video of a cat with mixed black and white fur, walking in the middle of an open roadway, carrying a cardboard sign that says "Meow I'm a Cat". In the distance behind is a green road sign that says "Model Testing Street" ran on each model.
  • For all models, "standard parameters" are used.
    • Steps is set to 20 for all models.
    • Frame count is set as model default.
    • CFG is set appropriate to the model.
    • Resolution is model default.
    • FPS is model default.
    • Note that outputs are converted and shrunk to avoid wasting too much space / processor power on the docs page.
  • For image2video models, an era-appropriate text2image model is used and noted.
  • This is just the image test prompt from Model Support but I swapped 'photo' to 'video', 'sitting' to 'walking', and 'holding' to 'carrying'. Goal is to achieve the same test as the image prompt does, but with a request for motion.
  • All generations are done on the base model of the relevant class, not on any finetune/lora/etc. Finetunes are likely to significantly change the qualitative capabilities, but unlikely to significantly change general ability to understand and follow prompts.
  • At time of writing, Hunyuan Video is the only properly good model. LTXV is really fast though.

Basic Usage

There's a full step by step guide for video model usage here: #716

Text-To-Video Models

  • Select the video model in the usual Models sub-tab, and configure parameters as usual, and hit Generate.
  • The Text To Video parameter group will be available to configure video-specific parameters.

Image-To-Video Models

  • Select a normal model as the base in the Models sub-tab, not your video model. Eg SDXL or Flux.
  • Select the video model under the Image To Video parameter group.
  • Generate as normal - the image model will generate an image, then the video model will turn it into a video.
  • If you want a raw/external image as your input:
    • Use the Init Image parameter group, upload your image there
    • Set Init Image Creativity to 0
    • The image model will be skipped entirely
    • You can use the Res button next to your image to copy the resolution in (otherwise your image may be stretched or squished)

Video Models

Hunyuan Video

hunyuan-video

This section is for the original Hunyuan Video (v1), for later version see next major section below.

Hunyuan Video Basic Install

Hunyuan Video Parameters

  • Resolution: The model is trained for 1280x720 (960x960) or 960x544 (720x720) resolutions or other aspect ratios of the same total pixel count
    • Using a lower resolution, like 848x480, can work with only some quality loss, and much lower mem/gen time.
  • FPS: The model is trained for 24 fps (cannot be changed, editing the FPS value will just give you 'slowmo' outputs)
  • FrameCount (Length): The model supports dynamic frame counts (eg 73 or 129 is solid), so you can pick the duration you want via the Text2Video Frames parameter.
    • Multiples of 4 plus 1 (4, 9, 13, 17, ...) are required due to the 4x temporal compression in the Hunyuan VAE.
    • The input parameter will automatically round if you enter an invalid value.
    • For quick generations, 25 is a good short frame count that creates about 1 second of video.
      • Use 49 for 2 seconds, 73 for 3 seconds, 97 for 4 seconds, 121 for 5 seconds, 145 for 6 seconds,
    • Supposedly, a frame count of 201 yields a perfect looping video (about 8.5 seconds long).
  • Guidance Scale: Hunyuan Video is based on the Flux Dev architecture, and has similar requirements.
    • Set the core CFG Scale parameter to 1.
    • You can use the Flux Guidance Scale parameter on this model (for Hunyuan Video, unlike Flux Dev, this value is embedded from CFG scale, and so prefers values around 6).
      • For "FastVideo" raise it up to 10.
  • Sigma Shift: Leave Sigma Shift disabled for regular Hunyuan Video, but for "FastVideo" enable it and raise it to 17.

Hunyuan Video Performance / Optimization

  • Hunyuan Video is very GPU and memory intensive, especially the VAE
    • Even on an RTX 4090, this will max out your VRAM and will be very slow to generate. (the GGUF models help reduce this)
  • The VAE has a harsh memory requirement that may limit you from high duration videos.
    • VAE Tiling is basically mandatory for consumer GPUs. You can configure both image space tiling, and video frame tiling, with the parameters under Advanced Sampling.
    • If you do not manually enable VAE Tiling, Swarm will automatically enable it at 256 with 64 overlap, and temporal 32 frames with 4 overlap. (Because the memory requirements without tiling are basically impossible. You can set the tiling values very very high if you want to make the tile artifacts invisible and you have enough memory to handle it).
  • By default the BF16 version of the model will be loaded in FP8. To change this, use the Preferred DType advanced parameter.
    • FP8 noticeably changes results compared to BF16, but lets it run much much faster.
    • The GGUF versions of the model are highly recommended, as they get much closer to original and very close performance to fp8.
      • GGUF Q6_K is nearly identical to BF16.

Hunyuan Video Additional Notes

  • You can use Hunyuan Video as a Text2Image model by setting Text2Video Frames to 1.
    • The base model as an image generator performs like a slightly dumber version of Flux Dev.

FastVideo

  • FastVideo is a version of Hunyuan Video trained for lower step counts (as low as 6)
  • You can get the FastVideo fp8 from Kijai https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_FastVideo_720_fp8_e4m3fn.safetensors
    • Save to the diffusion_models folder
  • Or the gguf FastVideo from city96 https://huggingface.co/city96/FastHunyuan-gguf/tree/main
    • Save to the diffusion_models folder, then load up Swarm and click the ☰ hamburger menu on the model, then Edit Metadata, and set the Architecture: field to Hunyuan Video (this might autodetect but not guaranteed so double-check it)
  • Set the advanced Sigma Shift param to a high value around 17
  • Set the Flux Guidance at a higher than normal value as well (eg 10).
  • Not adjusting these values well will yield terribly distorted results. Swarm does not automate these for FastVideo currently!

Hunyuan Image2Video

  • Hunyuan Image2Video is the official image-to-video model from Hunyuan's team, install info above.
  • Works like any other Image2Video model, with the same general parameter expectations as regular Hunyuan Video.
  • For I2V "v1", You will want to use the Advanced -> Other Fixes -> Trim Video Start Frames parameter with a value of 4, as the model tends to corrupt the first few frames.
  • For I2V "v2" / "Fixed" version, you will need to click the ☰ hamburger menu on the model, then Edit Metadata, and set the Architecture: field to Hunyuan Video - Image2Video V2 ('Fixed')

SkyReels Text2Video

  • SkyReels is a finetune of Hunyuan video produced by SkyWorkAI, see their repo here
  • You can download a SkyReels Text2Video fp8 model from here https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/blob/main/skyreels_hunyuan_t2v_fp8_e4m3fn.safetensors
    • Save to the diffusion_models folder
  • Broadly used like any other Hunyuan Video model
  • This model prefers you use real CFG Scale of 6, and set Flux Guidance value to 1
  • Their docs say you should prefix prompts with FPS-24, as this was trained in. In practice the differences seem to be minor.
  • Sigma Shift default value is 7, you do not need to edit it

SkyReels Image2Video

  • You can download a SkyReels Image2Video fp8 model from here https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/blob/main/skyreels_hunyuan_i2v_fp8_e4m3fn.safetensors
    • Save to the diffusion_models folder
  • Or you can select a gguf variant from https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/tree/main
    • Save to the diffusion_models folder, reload Swarm models list, click the ☰ hamburger menu on the model, then Edit Metadata, and set the Architecture: field to Hunyuan Video - SkyReels Image2Video
  • Use via the Image To Video param group
  • This model prefers you use real CFG Scale of around 4 to 6, and set Flux Guidance value to 1
  • Their docs say you should prefix prompts with FPS-24, as this was trained in. In practice the differences seem to be minor.
  • The model seems to be pretty hit-or-miss as to whether it creates a video of your image, or just "transitions" from your image to something else based on the prompt.
  • The model seems to have visual quality artifacts
    • Set Video Steps higher, at least 30, to reduce these
  • Sigma Shift default value is 7, you do not need to edit it

Hunyuan Video 1.5

hyvid15-cattest.mp4

(Hunyuan Video 1.5 - T2V 720p non-distilled CFG=6 Steps=20 Frames=121)

  • SwarmUI supports Hunyuan Video 1.5 Models
    • There appear to be quality issues not related to the Swarm impl, either in the model or in the upstream comfy impl.
  • Downloads here https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged/tree/main/split_files/diffusion_models
    • save to diffusion_models folder
    • There are variants for Text2Video vs Image2Video
      • Despite the labeled difference, both variants can equally do both text2video and image2video.
      • Also a dedicated superresolution v2v upscaler, see below
    • There are 480p and 720p variants
      • Swarm will assume all models are 720p (960x960). For the 480p models, you may want to edit the model metadata and set the resolution to 640x640.
      • They are actually pretty friendly to mixing the resolution, 720p can do 480 fine and 480p can mostly do 720.
    • There are CFG Distilled and non-distilled versions
      • CFG distilled runs faster and with less vram, non-distilled is slower but MIGHT yield better quality
  • The VAE is a 16x16 downsample (as opposed to most prior models using 8x8)
    • This allows HyVid1.5 to run faster than most, but with some quality reduction on small details
  • Parameters:
    • CFG: 1 for Distilled, otherwise normal high CFG values, eg 6
    • Steps: Normal step counts (20+)
    • Frames: Trained for 121 (5 seconds), shorter lengths work fine too, or longer up to 241 (10 seconds). When not specified, Swarm will default to 73 (3 seconds).
      • You can do Frames=1 for image generation.
    • FPS: The model is trained for 24 fps
    • Resolution: Aside from the trained resolution, the models seem happy with different smaller resolutions or different aspect ratios as well.
    • Sigma Shift: defaults to 7. They recommend lowering to 5 for 480p and raising to 9 on specifically 720p T2V.
      • SuperResolution uses 2

Hunyuan Video 1.5 SuperResolution Model

  • The SuperResolution models function equivalent to basic models, and are meant to be used as a Refiner model.

    • Save in the same folder as the rest.
    • You may need to manually edit the model metadata to architecture Hunyuan Video 1.5 SuperResolution
    • Probably edit model metadata to set the resolution to 1920x1080 (or approx 1:1 of 1456x1456)
    • The SR models have "distilled" in the filename but seem to respond better to CFG=6 and make a mess at CFG=1.

  • There are dedicated latent upscale models here https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged/tree/main/split_files/latent_upscale_models

    • Save to (SwarmUI)/Models/latent_upscale_models (create the folder if it doesn't already exist)
    • Select the model as the Refiner Upscale Method
    • If you have a 720p gen, and you are using the 1080p upscale, set Refiner Upscale to 1.5.

  • Not yet supported, WIP.

Lightricks LTX Video

ltxv

(LTX-Video 0.9.1, Text2Video, CFG=7 because 3 was really bad)

LTXV Install

  • Lightricks LTX Video ("LTXV") is supported natively in SwarmUI as a Text-To-Video and also as an Image-To-Video model.
    • The text2video is not great quality compared to other models, but the image2video functionality is popular.
  • Download your preferred safetensors version from https://huggingface.co/Lightricks/LTX-Video/tree/main
    • At time of writing, they have 0.9, 0.9.1, 0.9.5, and 0.9.6, each new version better than the last, but all pretty bad
    • save to Stable-Diffusion folder
    • The text encoder (T5-XXL) and VAE will be automatically downloaded
      • You can also set these manually if preferred
  • On the Server -> Extensions tab, you'll want to grab SkipLayerGuidanceExtension, so you can use "STG", a quality improvement for LTXV

LTXV Parameters

  • FPS: The model is trained for 24 fps but supports custom fps values
  • Frames: frame counts dynamic anywhere up to 257. Multiples of 8 plus 1 (9, 17, 25, 33, 41, ...) are required due to the 8x temporal compression in the LTXV VAE. The input parameter will automatically round if you enter an invalid value.
  • Resolution: They recommend 768x512, which is a 3:2 resolution. Other aspect ratios are fine, but the recommended resolution does appear to yield better quality.
  • CFG: Recommended CFG=3
  • Prompt: very very long descriptive prompts.
    • Seriously this model will make a mess with short prompts. If you ask for a video of a cat you will just get a dark blur.
    • Example prompt (from ComfyUI's reference workflow):
      • Prompt: best quality, 4k, HDR, a tracking shot of a beautiful scene of the sea waves on the beach
      • Or Prompt: A drone quickly rises through a bank of morning fog, revealing a pristine alpine lake surrounded by snow-capped mountains. The camera glides forward over the glassy water, capturing perfect reflections of the peaks. As it continues, the perspective shifts to reveal a lone wooden cabin with a curl of smoke from its chimney, nestled among tall pines at the lake's edge. The final shot tracks upward rapidly, transitioning from intimate to epic as the full mountain range comes into view, bathed in the golden light of sunrise breaking through scattered clouds.
      • Negative Prompt: low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly
  • If you installed the SkipLayerGuidanceExtension, Find the Skip Layer Guidance parameter group in advanced
    • Set [SLG] Scale to 1
    • Leave Rescaling Scale and Layer Target unchecked, leave the start/end percents default
  • LTX has some official tips and info on their HF page https://huggingface.co/Lightricks/LTX-Video

LTXV Image To Video

  • You can use the regular LTXV model as an Image-To-Video model
    • Select the LTXV model under the Image To Video group's Video Model parameter
    • Set Video FPS to 24 and Video CFG to 3, set Video Frames to a higher value eg 97
    • Pay attention that your prompt is used for both the image, and video stages
      • You may wish to generate the image once, then do the video separately
      • To do that, set the image as an Init Image, and set Creativity to 0

LTXV Performance

  • LTXV has the best performance of any video model supported in Swarm. It is wildly fast. This comes at the cost of quality.

Lightricks LTX Video 2

  • LTX-2 is the first proper Audio+Video combo model available as open source
    • Also known as "LTXV2", "LTXAV", ...
  • SwarmUI has basic support for LTXV-2 (however the model is new and has very different tech than usual, so some edge cases are weird)
    • Download the model from Lightricks
    • Save in Stable-Diffusion models folder
    • More work to be done still on fully supporting the extent of model capabilities
  • LTXV-2 has a dedicated latent spatial upscaler model
    • If you want to use it, download ltx-2-spatial-upscaler-x2-1.0.safetensors
    • save it to (SwarmUI)/Models/latent_upscale_models
    • Set Refiner Upscale to 2, select the model as the Refiner Upscale Method parameter, and set Refiner Control Percentage to 0.5. Set your base resolution to half of your target (eg 320 instead of 640).
      • The upscaler is hardlocked at 2x and will not work at any other upscale amount.
    • This is for T2V only.
    • It seems largely redundant, the model works about the same if you just don't bother using this.
  • Parameters:
    • Prompt: LTXV really needs long prompts to accomplish anything.
      • They have an official prompting guide but it only covers T2V
      • I2V prompting is more akin to Wan/other model i2v prompting: describe what actions to take, what to change, what to add. Avoid redescribing the scene already present in the image.
      • LLM prompt rewrite may be necessary to get the most out of the model.
    • CFG Scale: The regular model uses normal CFG values such as ~4, the distilled model uses 1.
    • Steps: The regular model uses normal step values, 20+. The distilled model uses 8 but works at 4.
    • Frames: 8n+1 (9, 17, 25, 33, 41, 49, ...)
    • Sampler: Defaults to Euler
    • Scheduler: Defaults to Normal. With an upscale/refiner, Linear Quadratic may be better
    • Negative Prompt: Reference workflow suggests using this giant negative:
      Giant negative blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, wrong hand count, artifacts around text, unreadable text on shirt or hat, incorrect lettering on cap (“PNTR”), incorrect t-shirt slogan (“JUST DO IT”), missing microphone, misplaced microphone, inconsistent perspective, camera shake, incorrect depth of field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, smiling, laughing, exaggerated sadness, wrong gaze direction, eyes looking at camera, mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, off-sync audio, missing sniff sounds, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, missing door or shelves, missing shallow depth of field, flat lighting, inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts.

LTX-2.3

Wan 2.1

wan21_14b

(Wan 2.1 - 14B Text2Video)

wan21_13b

(Wan 2.1 - 1.3B Text2Video)

Wan 2.1 Install

Wan 2.1 Parameters

  • Prompt: Standard. Supports English and Chinese text.
    • They have an official reference negative prompt in Chinese, it is not required but may help: 色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走
      • (This is just a word spam negative "bright colors, overexposed, static, blurred details, subtitles, ..." but in Chinese. It does help though.)
  • FPS: The original Wan 2.1 base model is trained for 16 FPS. Most variants, including Wan 2.2-5B, CausVid, Lightx2v, Lightning, etc, are trained for 24 FPS.
    • Swarm will default to 24 FPS for Wan. You must manually select 16 FPS when using the original Wan 2.1 base.
  • Resolution: The models are trained for 832x480, which is a 16:9 equivalent for 640x640
    • the 14B models are trained also for 1280x720, which is a 16:9 equivalent for 960x960
    • Other resolutions seem to work fine. Even the 1.3B, which is not trained for 960, can technically still do 960 just with a quality drop as it gets too large.
      • As a vid2vid gen, the model seem to be very good at generating very high res directly.
    • Any aspect ratio is fine.
  • Frame Count (Length): you can select pretty freely, different values work fine. If unspecified, will default to 81 (5 seconds if at 16 fps).
    • Use 17 for one second, 33 for two, 49 for three, 65 for 4, 81 for 5.
    • Higher frame counts above 81 seem to become distorted - still work but quality degrades and glitching appears.
    • The Text2Video models seem to favor 81 frames (5 seconds) and exhibit some signs of quality degradation at very low values, the Image2Video models are much more malleable
  • Steps: Standard, eg Steps=20, is fine. Changing this value works broadly as expected with other models.
    • Slightly higher (25 or 30) is probably better for small detail quality
  • CFG Scale: Standard CFG ranges are fine. Official recommended CFG is 6, but you can play with it.
    • Image2Video models may work better at lower CFGs, eg 4. High CFGs will produce aggressive shifts in lighting.
  • Sampler and Scheduler: Standard, eg Euler + Simple
    • You can experiment with changing these around, some may be better than others
  • Sigma Shift: range of 8 to 12 suggested. Default is 8.
  • Performance:
    • Wan 14B is pretty slow unless you have top-end hardware (4090/5090). With topend hardware and all the best speed optimizations... it's still generally going to run in the "minutes per video" range.
    • If you see generations completing but then freezing or dying at the end, the advanced VAE Tiling parameters may help fix that. Ignore the temporal tiling (the Wan VAE is implicitly temporally tiled).
    • The Image2Video models are much more performance-intensive than the Text2Video models
    • The lightning/causvid/lightx2v models make Wan much faster, see the section below
    • To run faster, use a "HighRes Fix" style setup, there's a guide to that here: https://www.reddit.com/r/StableDiffusion/comments/1j0znur/run_wan_faster_highres_fix_in_2025/
  • Quality:
    • The Wan models sometimes produce glitched content on the first or last few frames - under Advanced->Other Fixes->you can adjust Trim Video Start Frames (and End) to a small number (1 to 4) to cut the first/last few frames to dodge this.

Wan CausVid - High Speed 14B

Wan For Image Generation

  • You can use Wan T2V as an image generation model too!
  • Just set Text2Video Frames to 1
  • This works for all Wan T2V variants (2.1 1.3B, 2.1 14B, 2.2 14B, 2.2 5B, ...)
  • This is compatible with Lightx2v/Lightning LoRAs.
  • Some parameter adjustments may be needed
    • Notably, setting Sigma Shift to 1 or 2 seems to improve quality significantly.
    • Wan may be overly resolution and aspect sensitive when generating images
      • For example, 3:4 or 2:3 at side length 1280 might make pretty great portraits on some Wan variants, but swap to 9:16 or to default side length on the same model and it looks terrible.

Wan Phantom

Wan VACE

  • Wan VACE has initial support in SwarmUI.
    • For Reference Image mode:
      • Select your VACE model as your regular text2video base model
      • Set the reference image as your Init Image
      • Set Creativity to 1
      • Generate as normal
    • For Control Video mode:
      • Not yet supported in main interface, use Comfy Workflow tab

Wan 2.2

  • Wan 2.2 is natively supported in SwarmUI
    • Wan 2.2 is better in some regards (notably photorealistic video quality), but not all, compared to Wan 2.1. Notably, it is more complicated to set up. If you're new, start with Wan 2.1, and try an upgrade to 2.2 after you're familiar with the basics.
    • You can download the standard version of the model(s) from here https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/diffusion_models
      • Or, there's a collection of GGUF files here: https://huggingface.co/collections/QuantStack/wan22-ggufs-6887ec891bdea453a35b95f3
      • There's a 14B T2V (Text To Video), in a high+low noise pair
        • You're expected to run the high noise as a base and the low noise as a refiner, with:
          • RefinerMethod as StepSwap, and
          • RefinerControlPercentage as 0.5 (or higher if preferred, cannot go lower)
        • Reference CFG range is 5
      • There's a 14B I2V (Image To Video), in a high+low noise pair
        • You're expected to run the high noise as a base and the low noise as a refiner
          • In the Image To Video params:
            • Set the regular Video Model to the high noise model,
            • and set the advanced Video Swap Model to the low noise model,
            • and leave Video Swap Percent at 0.5 (or higher if preferred, cannot go lower)
        • Reference CFG range is 3.5
        • This also supports the Video End Frame input to create a video that moves between two known places
        • The "Low" is very close to the original Wan 2.1, and can be ran as 2.1 just fine. "High" is a much further retune that can only do the initial ~50% of steps properly.
      • For both 14B types:
        • FPS is 16, but loras or even parameter adjustments can change it to a more normal-looking 24.
          • Swarm will default to 24, but if your videos "feel sped up", change the FPS parameter to 16.
        • Sigma shift may be worth experimenting with. The default is 8, but a wide range of values are functional.
          • Some users recommend 1.5 for T2V
      • There's a 5B T/I2V (single model that does both text and image to video) as well
        • It has its own VAE. Will be autodownloaded.
        • No funky model pair like the 14b has, just a straight single model
        • Reference CFG is 3.5
        • Native FPS of 24
      • There are some Wan 2.2 Lightx2v models available
        • Notably this pair: https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22-Lightning
        • You use a separate High and Low variant together
        • There are two ways to set the pair loras...
          • Option 1: via the UI
            • Click "Edit Metadata" on the model, find "Default Confinement", select the appropriate confinement, and hit Save
              • For T2V high this is "Base", for T2V Low this is "Refiner", for I2V High this is "Video", for I2V Low this is "VideoSwap"
            • Then just select both loras as normal
          • Option 2: in the prompt
            • For T2V Use with this at the end of your prompt: <base> <lora:Wan2.2-Lightning_T2V-A14B-4steps-lora_HIGH_fp16> <refiner> <lora:Wan2.2-Lightning_T2V-A14B-4steps-lora_LOW_fp16> (adapt the lora filenames to whatever specific filenames you have locally)
              • You can use your LoRA browser tab at the top, find the LoRA, and click the ☰ hamburger menu and then Add To Prompt
            • For I2V, use <video> <lora:...i2v-high> <videoswap> <lora:...i2v-low> and of course use the i2v loras
              • the I2V Lightning LoRA appears to target 16 fps
            • Because this is wonky, once you get it working, it is recommended that you make a Preset with the Prompt set like {value} <base> ... <refiner> ... to make it easy to click straight into this behavior rather than doing it manually every time. You can also select the models, CFG, etc. in the preset to have it all ready in one click.
      • You can use the Wan 2.1 Lightx2v or other causvid-likes (see CausVid Section Above) on the Wan 2.2 14B (not on the 5B)
        • For I2V, this seems to "just work"
        • For T2V, this has some visual oddities but does still mostly work
      • Wan 2.2 has an official prompting guide book: https://alidocs.dingtalk.com/i/nodes/EpGBa2Lm8aZxe5myC99MelA2WgN7R35y

Kandinsky 5

  • Kandinsky 5 Video Lite and Video Pro are supported in SwarmUI!
  • They come in a variety of variants, you will have to pick what you want, or experimental with several.
  • At time of writing, the current implementation has bugs, and some hacks are used to workaround them. Not all features work. What does work is kinda bad.
  • Parameters:
    • These vary heavily based on model you choose.
    • CFG Scale: for regular models, regular CFG such as 5 works. For CFG-distill and step distill, use CFG of 1.
    • Steps: For regular, 20 or higher is used. For Step Distill, 16 is the target. Going lower will work but with a severe quality hit.
    • Resolution: All video models primarily target a side length of 640. Higher resolutions can work, Pro handles 960x960 fine.

Obscure Model Redirection

Stable Video Diffusion

Genmo Mochi 1 (Text2Video)

NVIDIA Cosmos

These obscure/old/bad/unpopular/etc. models have been moved to Obscure Model Support