---
license: apache-2.0
base_model: Wan-AI/Wan2.1-T2V-14B
tags:
  - wan
  - video
  - text-to-video
  - diffusion-pipe
  - lora
  - template:sd-lora
  - standard
library_name: diffusers
pipeline_tag: text-to-video
instance_prompt: Jodorowsky psychedelic montage 1970s film by jodorowsky
widget:
- text: >-
    Wan2.1 LoRA:Run3@2950
  output:
    url: samples/1755569403645__000002950_0.webp
- text: >-
    Wan2.1 LoRA:Run3@3450
  output:
    url: 3450_Run3_ii.mov
- text: >-
    Wan2.1 LoRA:Run3@2750
  output:
    url: samples/1755566258585__000002750_0.webp
- text: >-
    Wan2.1 LoRA:Run2@1600
  output:
    url: samples/1755545990540__000001600_0.webp
- text: >-
    Wan2.1 LoRA:Run1@1000
  output:
    url: 1000_Wan21i.mov
- text: >-
    Wan2.1 LoRA:Run3@2450
  output:
    url: samples/1755561576004__000002450_0.webp
- text: >-
    Wan2.1 LoRA:Run3@3450
  output:
    url: 3450_Run3_i.mov
- text: >-
    Wan2.1 LoRA:Run3@2350
  output:
    url: samples/1755560004870__000002350_0.webp
- text: >-
    Wan2.1 LoRA:Run2_pEMA_SigmaRel0.19
  output:
    url: 2000pEMA019_Run2_i.mov
- text: >-
    Wan2.2 LoRA:Run1@1000
  output:
    url: Wan22_St1000_jodorowsky1.mov
---
## ALEJANDRO JODOROWSKY's CINE-SURREELS
## A Low(ish) Rank Adapter (LoRA) 
## For Wan2.* 14B Text to Video Models
## ____||| By SilverAgePoets.com |||____

Artistically-specialized text to video generative fine-tuned low-rank adapter (Rank 16 LoRA) for the 14billion-parameter Wan2.1, Wan2.2, and derived base models. <br>
This LoRA was trained on a custom dataset of video clips from classic films by **Alejandro Jodorowsky**: the great filmmaker, artist, author, psychoanalyst, sage, & occultist/psyche-mage... <br>
<Gallery />

## To reinforce the adapter, pre-phrase/amend prompts with:
`[Jodorowsky] psychedelic montage 1970s film by Alejandro Jodorowsky`, etc... <br>
Other suggested prompt-charms: `surrealist occult cinema, eclectically collaged scene, dynamic motion, kodachrome, classic countercultural movie, experimental arthouse analog footage`, etc... <br>

## Training/Usage Notes:
The training, orchestrated using Ostris' [ai-toolkit trainer](https://github.com/ostris/ai-toolkit/tree/main), was conducted in several stages/runs, with each pause/re-start involving a partial changing-out of trained-on clips and substantial modifications of hyperparameters: <br> 
**Run 1**: Steps 0 thru 1000. With lr: 1e-4, `content_or_style: content` (high noise stage emphasis), and medium resolution samples.  <br>
**Run 2**: Steps 1001 thru 1800. With `content_or_style: balanced` (balanced noise schedule), lr: 9e-5, lower resolution and changed out/more numerous samples.  <br>
**Run 3**: Steps 1801 thru 3450. With higher resolution samples than previous runs, plus an additional (to linear) training of `conv` & `conv_alpha` networks at rank 16, with `force_consistent_noise: True`, and `content_or_style: content` (high noise stage emphasis), and lr: 1e-4.  <br>
This adapter works with both Wan2.1 and Wan2.2. Euler schedulers work best in our tests. Lower "shift" values typically yield more realism/analog quality, depending on other factors. <br>

## With accelerated Inference LoRAs:
All checkpoints are confirmed to work with the Wan2.1 Self-Forcing (T2V), FastWan, and CauseVid accelerater adapter LoRAs. <br>
Checkpoints from Run 3 (but not Runs 1 or 2) are confirmed to work with the Wan2.2 T2V Lightning/4-step Adapter (for the Low Noise Expert transformer). <br>
All checkpoints are also likely to work with the Wan2.2 High Noise T2V Lightning/4-step Adapter, but at worse quality/detailing/chromatic range. <br>