Transfer from
https://huggingface.co/eddy1111111/WAN22.XX_Palingenesis/tree/main
Better motion effects and closer adherence to prompts; works best when paired with a specific textEncode T5 model. Link:
You can’t get good results without referring to this workflow:
https://www.runninghub.ai/post/1973349038638673921
Use the prompts below to compare with other versions of the model:
• Core Elements: Subject (Blonde, Topless) + Setting (Spring Bedroom, Wooden Desk + White Gauze Curtains) + Movement (Book Turning / Hair Fluttering)
• Aesthetic Controls: Light Source (Daylight, Side Light Through Gauze Curtains), Light Type (Soft Light), Shot Size (Medium Close-Up), Composition (Center Composition), Color Tone (Warm, Low-Saturation)
• 5-Second Action Breakdown:
0-2 seconds: The woman holds the book page with her fingertips and gently turns it to the right, a gentle breeze blowing through the pages;
2-4 seconds: She looks out the window, her eyes soft, as the gauze curtains brush against her shoulders in the wind;
4-5 seconds: Her hair sways gently in the breeze as she lowers her head to continue reading the book, a slight smile on her lips.
********************************************************
• Core elements: subject (nude blonde woman) + setting (city street at dusk, graffiti wall + asphalt road) + motion (pushing off to accelerate / aerial tuck jump)
• Aesthetic controls: light source (dusk light, side lighting), light type (hard light, pronounced shadows), framing (mid-shot), camera movement (camera follows, slight orbit), color tone (warm tones, graffiti wall highly saturated)
• 5-second action breakdown:
0–2 s: the nude woman pushes off with her left foot, the skateboard accelerates forward, her body leans slightly forward;
2–3 s: both feet tap the board simultaneously, the skateboard pops into the air and flips, her knees bend and pull in;
3–5 s: the skateboard lands and rolls, the nude woman shifts slightly to adjust her center of gravity, arms hanging naturally to maintain balance.
********************************************************
• Core elements: subject (a nude blonde woman with translucent wings on her back) + setting (a magical forest at dawn, moss-covered stones + unopened flower buds) + motion (reaching out to cast a spell / petals unfurling)
• Aesthetic controls: light source (dawn light, soft light), light type (rim light, outlining the fairy wings), shot size (medium-close), composition (central composition), color tone (warm tones, petals pale pink-white + wing reflections)
• 5-second action breakdown:
0–2 s: She crouches before a bud, palms facing the bud, a faint green glow emanating from her hands;
2–4 s: The bud slowly opens from the top, pink petals curling upward, the yellow stamen of the flower revealed;
4–5 s: Dew on the petals refracts the morning light, she lightly touches the flower’s center, and her wings flutter gently.
Description
FAQ
Comments (4)
Please describe what precisely is improved and how here, I've never heard of this model; there's no model card; where did this come from, what was changed? The example doesn't show much improvement.
Does it still require Lightning Loras to reduce step count or is this built in? The opening sentence describing the Palingenesis model makes it sound like it might already have this baked in... I did check the linked repo, but helpfully there is no model page to identify anything about the model (as seems extremely common on HF!)
is this the same as "WAN22.XX_Palingenesis_high_i2v_fix.safetensors" or have you done any changes?