Creating convincing multi-performer AI porn has always been one of the toughest challenges in AI video generation. Single-performer scenes are relatively straightforward, but the moment you add a second person the model has to track two bodies, coordinate their movements, and keep the action physically coherent. If you've been experimenting with making AI porn, you already know that complex scenes with multiple performers tend to fall apart fast. The Double Blowjob workflow solves this by splitting the problem in two.

The core issue is simple: no single LoRA can handle everything from the approach to the action. A Double BJ LoRA trained specifically on two women sucking one cock can nail the oral action itself, but it has no idea how to get them from standing to kneeling. The moment you try to prompt the full sequence — approach, kneel, start licking — in one generation, the model gets confused and produces garbage. Hands merge into faces, bodies clip through each other, and the spatial relationship between all three people collapses.

The Technical Breakdown

• 2-pass Image-to-Video workflow — each pass handles one distinct phase of the action
• Pass 1: W22_Multiscene_BJ LoRA (strength 1.0) manages the approach and kneeling transition across 61 frames. This is the proven BJ NOW LoRA that reliably handles standing-to-kneeling movement.
• Pass 2: WAN-2.2-I2V-Double-Blowjob LoRA (strength 1.0) takes over with both women already in position, generating 101 frames of dual licking, sucking, and sharing.
• Checkpoint: DasiwaWAN22I2V14B (High + Low) for both passes
• Both passes use WanImageToVideo — the last frame of Pass 1 becomes the starting image for Pass 2
• Post-processing: passes merged via VHS_MergeImages, upscaled 2x, RIFE interpolation 2x, rendered at 40fps
• Total output: approximately 13 seconds of seamless video

The key innovation here is the handoff between LoRAs. The Double BJ LoRA was only trained on the action itself — both women already positioned and working the cock. It cannot handle the standing-to-kneeling transition at all. So Pass 1 uses the proven Multiscene BJ LoRA to get both women down on their knees and looking up at the camera. The last frame of that generation, with both women in perfect position, becomes the starting image for Pass 2. The Double BJ LoRA then takes over from that position and handles the actual oral action with full confidence.

Workflow Specs

• Sampler: lcm
• Scheduler: beta
• Steps: 6
• CFG: 1
• Pass 1 frames: 61
• Pass 2 frames: 101
• Pass 1 LoRA: W22_Multiscene_BJ (1.0)
• Pass 2 LoRA: WAN-2.2-I2V-Double-Blowjob (1.0)
• Upscale: 2x
• Frame interpolation: RIFE 2x
• Output FPS: 40
• Input: Upload an image of two women together

The prompt drives the full sequence: “Both women approach and go down to the floor, on their knees and then look up at the camera. One woman grasps the penis with her hands, moves her hands up and down, licks the penis and puts it inside her mouth. The other woman licks the shaft from the side. They take turns sucking and licking, sharing the cock between them.”

Related Workflows

• WAN 2.2 NSFW Workflows — Full Overview
• WAN 2.2 Blowjob & Deepthroat Workflow
• WAN 2.2 Smooth Sex Workflow
• WAN 2.2 Anal Insertion Workflow

How to Use This Workflow

1. Prepare your input image. Generate or select a photo of two women standing or sitting together. Both faces should be clearly visible and well-lit. The image quality directly determines the output quality.
2. Run the workflow as-is. The two passes are pre-configured to chain automatically. Pass 1 handles the approach and kneeling, and its last frame feeds directly into Pass 2 for the oral action. You do not need to manually extract frames between passes.
3. Do not modify LoRA strengths. Both LoRAs run at 1.0 because each pass has only one LoRA doing one specific job. Lowering the strength introduces drift and the model loses track of the action.
4. Check your input image composition. Both women should be roughly the same distance from camera. If one is significantly closer or further away, the kneeling transition in Pass 1 tends to break spatial coherence.
5. Experiment with different starting poses. The workflow handles standing starts reliably, but seated or partially kneeling starting positions can produce interesting variations since Pass 1 has less transition work to do.

Ready to generate your own double blowjob scenes? Run this workflow in the VirtuaVixen AI Studio using your own images, or grab the full workflow file from the VirtuaVixen Pro ComfyUI Workflow Pack to run it locally on your own hardware.