WanGP v18.17 adds FLUX 3 Klein, Fish Audio Drama 3 and MOSS Image
WanGP v18.17 brings FLUX 3 Image Klein, Fish Audio Drama 3 dialogue with voice cloning and MOSS Image 5B to the low-VRAM local app. It also adds a lighter H3 video decoder, INT8 math kernels and a Deepy Prime agent that plans a one-minute video from a brief.
What it does
FLUX 3 Klein 5.6B covers text-to-image and editing with up to 10 reference images, distilled (16 steps) or Base (50 steps with guidance), in BF16, INT8 ConvRot and NVFP4, with a Prompt Helper for drawing layout boxes. Fish Audio Drama 3 writes speaker lines with emotions, whispers, laughs and pauses, clones up to three voices, and can generate long scripts in parts. MOSS Image 5B was added in the same update.
Deepy Prime takes a one-minute brief (an ad, a short film, a music video), plans the shots, keeps the cast consistent, paces the cuts and adds music or narration. It then checks for black frames, unwanted cuts, silent stretches and typos in on-screen text, and draws titles and lower thirds with exact lettering, with subtitles on a separate track.
Worth knowing
H3 LightVAE is a 12-layer decoder (BF16 or INT8) for H3's FL2VA and Ref2VA modes, and a new TaoMate H3 FL2VA 3-step profile was added. The Drama 3 INT4 Experts variant roughly halves the experts' memory. Layout boxes guide the composition but aren't exact pixel masks.
Try it yourself
In WanGP, pick Fish Audio Drama 3, add a short clean recording for each of [two] voices, and write: [Speaker 1, nervous] "[line]" (pause) [Speaker 2, laughing] "[line]".