"""16:9 night beach: Anima + charref keyframe, then the character frozen and the background redrawn (inverted WeVisOne matte).""" import json, sys, time, urllib.request from pathlib import Path sys.path.insert(0, str(Path(__file__).parent)) from gen_h3 import API, DATA, post, upload # noqa: E402 CHAR = "c0040" TAGS = "1girl, solo, blue hair, double bun, school uniform, serafuku, petite" KF_W, KF_H, W, H = 1344, 768, 832, 480 SCENE = "from behind, standing on the beach, looking at the horizon, night, starry sky, full moon, moonlight reflecting on the sea, waves, wide shot, scenery" MOTION = "waves rolling onto the shore, stars twinkling, moonlight shimmering on the water, wind" def graph(ref_image, prefix, denoise, grow, seed, noref=False, blur=1): neg = "worst quality, low quality, blurry, text, watermark" return { "1": {"class_type": "UNETLoader", "inputs": {"unet_name": "anima/anima-turbo-v1.0-int8-convrot.safetensors", "weight_dtype": "default"}}, "2": {"class_type": "CLIPLoader", "inputs": {"clip_name": "qwen_3_06b_base.safetensors", "type": "stable_diffusion", "device": "default"}}, "3": {"class_type": "VAELoader", "inputs": {"vae_name": "qwen_image_vae.safetensors"}}, "4": {"class_type": "LoadImage", "inputs": {"image": ref_image}}, "5": {"class_type": "AnimaRefAdapter", "inputs": {"model": ["1", 0], "vae": ["3", 0], "image": ["4", 0], "lora_name": "anima_ref_adapter_v2_r64.safetensors", "ref_guidance": 1.0, "correction": "none"}}, "6": {"class_type": "CLIPTextEncode", "inputs": {"clip": ["2", 0], "text": f"masterpiece, best quality, score_7, safe, {TAGS}, {SCENE}"}}, "7": {"class_type": "CLIPTextEncode", "inputs": {"clip": ["2", 0], "text": neg}}, # keyframe "10": {"class_type": "EmptyLatentImage", "inputs": {"width": KF_W, "height": KF_H, "batch_size": 1}}, "11": {"class_type": "KSampler", "inputs": {"model": ["5", 0], "positive": ["6", 0], "negative": ["7", 0], "latent_image": ["10", 0], "seed": seed, "steps": 8, "cfg": 1.0, "sampler_name": "euler", "scheduler": "simple", "denoise": 1.0}}, "12": {"class_type": "VAEDecode", "inputs": {"vae": ["3", 0], "samples": ["11", 0]}}, "13": {"class_type": "SaveImage", "inputs": {"images": ["12", 0], "filename_prefix": prefix + "_keyframe"}}, "14": {"class_type": "ImageScale", "inputs": {"image": ["12", 0], "upscale_method": "lanczos", "width": W, "height": H, "crop": "disabled"}}, # background-only redraw over 16 latent frames "20": {"class_type": "VAEEncode", "inputs": {"pixels": ["14", 0], "vae": ["3", 0]}}, **{str(i): {"class_type": "LatentConcat", "inputs": {"samples1": [str(i - 1), 0], "samples2": [str(i - 1), 0], "dim": "t"}} for i in range(21, 25)}, "25": {"class_type": "WeVisOneVLLoader", "inputs": {"clip_name": "wevisone_vl_4b_int8_convrot.safetensors"}}, "26": {"class_type": "WeVisOneVLPerceive", "inputs": {"clip": ["25", 0], "image": ["14", 0], "task": "matting", "query": "character", "max_length": 1024, "individual_masks": False}}, "27": {"class_type": "ThresholdMask", "inputs": {"mask": ["26", 3], "value": 0.3}}, "28": {"class_type": "GrowMask", "inputs": {"mask": ["27", 0], "expand": grow, "tapered_corners": True}}, # soft edge, so the frozen border does not keep a ring of the keyframe's background "40": {"class_type": "MaskToImage", "inputs": {"mask": ["28", 0]}}, "41": {"class_type": "ImageBlur", "inputs": {"image": ["40", 0], "blur_radius": blur, "sigma": blur / 2}}, "42": {"class_type": "ImageToMask", "inputs": {"image": ["41", 0], "channel": "red"}}, "29": {"class_type": "InvertMask", "inputs": {"mask": ["42", 0]}}, "30": {"class_type": "SetLatentNoiseMask", "inputs": {"samples": ["24", 0], "mask": ["29", 0]}}, # the background pass without the reference or character tags: with them the redrawn area grows another copy of her "31": {"class_type": "AnimaMotionModule", "inputs": {"model": ["1", 0] if noref else ["5", 0], "motion_name": "r2v_20clips/motion_02000.safetensors"}}, "32": {"class_type": "CLIPTextEncode", "inputs": {"clip": ["2", 0], "text": f"masterpiece, best quality, score_7, safe, {'no humans, ' if noref else TAGS + ', '}{SCENE.replace('from behind, standing on the beach, looking at the horizon, ', '') if noref else SCENE}, {MOTION}, anime screencap"}}, "33": {"class_type": "KSampler", "inputs": {"model": ["31", 0], "positive": ["32", 0], "negative": ["7", 0], "latent_image": ["30", 0], "seed": seed, "steps": 8, "cfg": 1.0, "sampler_name": "euler", "scheduler": "simple", "denoise": denoise}}, "34": {"class_type": "LatentCutToBatch", "inputs": {"samples": ["33", 0], "dim": "t", "slice_size": 1}}, "35": {"class_type": "VAEDecode", "inputs": {"vae": ["3", 0], "samples": ["34", 0]}}, "36": {"class_type": "CreateVideo", "inputs": {"images": ["35", 0], "fps": 8.0}}, "37": {"class_type": "SaveVideo", "inputs": {"video": ["36", 0], "filename_prefix": prefix, "format": "mp4", "codec": "h264"}}, "38": {"class_type": "MaskToImage", "inputs": {"mask": ["29", 0]}}, "39": {"class_type": "SaveImage", "inputs": {"images": ["38", 0], "filename_prefix": prefix + "_mask"}}, } def main(): ref = upload(DATA / CHAR / "ref.png") noref = "noref" in sys.argv[1:] grow = next((int(a[5:]) for a in sys.argv[1:] if a.startswith("grow=")), 16) blur = next((int(a[5:]) for a in sys.argv[1:] if a.startswith("blur=")), 1) for denoise in map(float, [a for a in sys.argv[1:] if a[0].isdigit()] or ["0.8"]): prefix = f"anima_motion_eval/beach{'_noref' if noref else ''}_d{int(denoise * 100)}_g{grow}b{blur}" pid = post("/prompt", json.dumps({"prompt": graph(ref, prefix, denoise, grow, 21, noref, blur)}).encode(), {"Content-Type": "application/json"})["prompt_id"] t = time.time() while pid not in (h := json.load(urllib.request.urlopen(f"{API}/history/{pid}"))): time.sleep(2) s = h[pid]["status"] print(prefix, s["status_str"], f"{time.time() - t:.0f}s", flush=True) if s["status_str"] != "success": print(json.dumps(s["messages"])[-1500:]) if __name__ == "__main__": main()