This model is based on https://civarchive.com/models/1878687/nsfw-image-to-video-by-wan22 with some fixes, simplification and additions.

Versions:

V1.1

• Added ability to set end frame from an image

T2V V1.0

• Added version for text to video with the same output for completness

• prompt is stolen from here https://civarchive.com/images/91061722

Instructions:

Step 1. Choose what input you want to use - uploaded image, stored video (probably produced by comfyui before) or uploaded video.

Step 2. Switch "Source Selector" accordingly.

Step 3. In a source group upload image/video or select existing one.

• if you selected video, last frame will be used as source. If you want to use different one edit "Select Images" node "indexes" field to value you want. -1 (currently set) means last frame, -2 the second from the end, 10 means 10th from begin. etc.

• Video will be produced from that frame but unlike automatic1111 will not be concatenated automatically, you need to do that in some video editor (I guess could be added in next version)

Step 4. Edit prompt.

Step 5. Set max side resolution (proportions will be kept), video duration, seed etc.

Step 6. Click "Run"

If everything works video file will be produced. If you want to change filename prefix, modify "Video Combine" node field "filename_prefix" in "Save Video" group. Also text file with the same name as video with extension ".txt" instead of ".mp4" will be generated, with main generation parameters stored.

Prompt is separated in 2 parts - "Prompt" and "Permanent Prompt". Last one is suppose to be quite permanent and does not need to change from video to video generation. Also clip duration added to "Permanent Prompt" automatically. So if it is "16fps, 4K, ultra-realistic skin texture, fabric details, seamless transition." after run it would be "Cinematic 8s clip, 16fps, 4K, ultra-realistic skin texture, fabric details, seamless transition."

Note: NSFW content can be generated with this workflow. Either switch to wan2.2 i2v checkpoint with NSFW content in "Wan Video Model Loader" nodes (high noise and low noise) (for example https://civarchive.com/models/2100616/wan-22-i2v-10-steps-nsfw-fp8-model ) and/or use LoRAs suitable for wan2.2 i2v in MultiLoras inputs.

Enjoy.