这次 H3 的偏差在时间：5.5 秒还没分开。 参考 @LiuLoky50572 的粉发、银发和烛光哥特氛围，我先用 imagegen 做了一张成年角色图，… · @superfang119 · TwKit

@superfang119

这次 H3 的偏差在时间：5.5 秒还没分开。 参考 @LiuLoky50572 的粉发、银发和烛光哥特氛围，我先用 imagegen 做了一张成年角色图，再交给本地 H3，生成8秒单镜头，只跑一版。 我写的是3—4秒一次闭唇轻吻，5.5秒前重新拉开。实际近接时男方嘴唇张开，闭唇要求没完全做到；5.5秒抽帧仍在接触，6.5秒才看到清楚间距。最后两人的眼神又接上了，黑蕾丝、刺绣和烛光也留住了；但不能说动作已经按秒完成。 我会留下这版的配色和结尾对视。要卡音乐节拍，接触与分开的时点还得修。这里只做了密集抽帧和关键帧细看，脸部接触、手指与衣料的全程关系没逐帧验，音轨也未主观试听。 这是新角色、新动作的主题改编，没有把原视频或音轨送进模型。下面两段都是实际使用的完整提示词，参数和来源一起公开。 Source / theme reference: @LiuLoky50572 https://x.com/LiuLoky50572/status/2111154176080462085 The source is a square video, approximately10.09s. Only a visible playback segment was inspected; the original full video and audio were not downloaded, used as model inputs or reproduced frame by frame. Pink/silver hair, gothic formalwear and candlelight inspired a newly created fully clothed adult couple. This is a theme adaptation with a new brief closed-lip kiss, not a copy of the source identities or explicit movement. The source author's complete execution prompt and model settings were not available in the visible post. Reference image: built-in imagegen, one newly generated1254x1254 image; no source image/video inputs. It is the sole Picture1 supplied to H3. The exact imagegen prompt is included below. Ref2VA provides image guidance, not exact first-frame or 3D geometry locking. Local ComfyUI / H3 Ref2VA; one generated video sample, no reroll, one image reference. Model: H3\minimax_h3_hybrid_fl2va_ref2va_b25-49-int8.safetensors Text/image encoder: h3\qwen3vl_32b_minimax_h3_ultra_uncensored_heretic_int8_convrot.safetensors LoRA: H3\minimax_h3_dmad_full_dareties_v4_step600_comfy_fro.safetensors x 0.80; H3\MysticXXX_MMH3-V1.safetensors x 0.45. SelfLift / Euler / simple, 12 steps, CFG1; lowres_scale0.6, transition_step6, rho0, w_min0.5/w_max1. Upscaler h3_upscaler_lms_v0.1.safetensors. Comfy kitchen attention + sol-attn (tau1.3, start0.2/end1); sigma shift video12/audio3. Negative conditioning: original ConditioningZeroOut. Canvas: 1:1 (Square), 1.5MP, original multiple32 -> 1248x1248, 24fps; requested8s, native192frames/8.000000s. Delivery is an unchanged native H.264/AAC copy, without cut, speed change or interpolation. Audio: newly generated AAC 32000Hz, 2 channels. No source audio was provided; generated audio has not been subjectively listened to. Video VAE: minimax_h3_video_vae_int8_convrot.safetensors Audio VAE: minimax_h3_audio_vae_fp32.safetensors Seed: 211115417608046. AI-generated image and H3 video. The H3 prompt is copied from the actual submitted workflow. Full actual imagegen prompt: Create one square cinematic dark-fantasy romantic portrait, a polished photorealistic still designed as a clear reference for a short film. Exactly two unmistakably adult fictional people, both about thirty years old, fully clothed in elegant opaque black gothic formalwear. At frame left, a woman with long pastel-pink hair, an elaborate black lace bow and a small silver drop earring; her high-neck black lace-trimmed dress has opaque lining and covers her chest and shoulders. At frame right, a man with tousled silver hair, a small silver earring, and a black embroidered high-collar jacket. Chest-up two-shot, the woman's face points right and the man's face points left, their eyes meeting warmly. Their faces are separated by a clearly visible small gap, lips closed, heads slightly tilted in complementary directions; neither face occludes the other. Her hands rest gently on the upper lapels of his jacket below his collarbone, separated from both faces. His hands remain below the frame. Relaxed affectionate expressions, natural adult facial proportions and skin texture, no exaggerated makeup. A dim gothic interior with dark carved wood and several distant candle bokeh lights; dusty rose rim light in the pink hair, warm amber candle highlights, deep charcoal shadows. Camera at eye level with a little space above the hair and around both profiles. Keep the staging readable and the two identities distinct, with no face fusion, extra person, text, watermark or split panel. This is a tender romantic film still: no tongue, no open-mouth contact, no nudity or sexualized pose. Full actual H3 prompt: subject_definitions: <Subject 1> is the fictional adult woman, about thirty years old, from <Picture 1>: long pastel-pink hair gathered with an intricate black lace bow, a small silver drop earring, natural adult facial proportions, and an opaque high-neck black gothic dress with long lace-trimmed sleeves. Keep her at frame left, facing right. <Subject 2> is the fictional adult man, about thirty years old, from <Picture 1>: tousled silver hair, a small silver earring, natural adult facial proportions, and an opaque embroidered black high-collar jacket. Keep him at frame right, facing left. <Subject 3> is the dark carved-wood interior and warm candle bokeh from <Picture 1>, with charcoal shadows and a dusty-rose highlight in the woman's hair. <Picture 1> is the newly created square, chest-up romantic two-shot. It guides identity, opaque clothing, lighting, relative positions and framing; it is an image reference rather than a required pixel-identical first frame. summary: [reference generation] Create an eight-second continuous chest-up shot of <Subject 1> and <Subject 2> in the candlelit setting of <Subject 3>. They move a small distance together, share one brief, tender closed-lip kiss, then separate to a clearly visible gap and look warmly at one another. Preserve the two distinct adult identities and the reference's fully clothed formal styling. The camera remains on the same side and almost still. All movement and audio are newly designed; no source video or source audio is a model input. retention_analysis: <Subject 1> (00:00.000-00:08.000): fully_preserved - maintain the woman's pink hair, lace bow, earring, high-neck opaque dress and left-screen placement while allowing the small head movement described below. Her two hands remain softly resting on the man's upper lapels, below his collarbone. <Subject 2> (00:00.000-00:08.000): fully_preserved - maintain the man's silver hair, earring, high collar and right-screen placement. His hands stay outside the lower frame throughout. <Subject 3> (00:00.000-00:08.000): fully_preserved - keep the same dim carved-wood room and the warm, out-of-focus candles. Only a subtle natural flame flicker changes the light. <Picture 1> (00:00.000-00:08.000): partially_preserved - use its chest-up square composition, identities, costume and palette as reference guidance; animate the new short interaction without claiming exact first-frame alignment or a locked three-dimensional scene. detailed_description: [Shot 1] (00:00.000-00:08.000), one uninterrupted medium close two-shot, square frame, eye-level camera. The adult woman is left and the adult man is right. Keep both heads and the top of her lace bow inside the image, both profiles readable, and the upper part of her hands visible on his jacket. Their opaque black clothes remain intact and cover the chest and shoulders. Soft amber candlelight catches cheek edges and embroidered fabric, pink hair has a muted rose rim, and carved wood remains a dark, softly blurred background. Use a virtually fixed camera with only an extremely slight, smooth push; no orbit, cut, reframing to the body, new person or lighting transition. From 00:00.000 to 00:01.500, begin with the faces clearly apart, looking into each other's eyes. Both lips are closed with small relaxed smiles. Her hands rest lightly on his lapels and do not climb toward his face or neck. A few fine hair strands and distant flames move subtly. Keep the visible separation between their two facial silhouettes. From 00:01.500 to 00:03.000, each leans forward a little. They tilt their heads gently in complementary directions so their noses pass without intersecting. Their eyelids lower naturally as the small gap closes. Keep the woman's pink hair and the man's silver hair distinct. Her fingers remain on the same fabric areas; do not introduce a new hand or move a hand across either face. From 00:03.000 to 00:04.000, they share exactly one brief, gentle closed-lip kiss. This is a restrained romantic film moment, without tongue, open-mouth contact or sexualized movement. Maintain two separate faces, natural noses and jawlines, with contact only at the closed lips; do not merge their heads, teeth, skin or hair. Hold the clothing and hand placement steady. From 00:04.000 to 00:05.500, they ease back a small distance and reopen a visible gap between their faces. They slowly open their eyes and resume eye contact, a slight smile at the corners of their mouths. Do not add a second kiss, a sudden head turn, a camera reversal or a new gesture. The woman's hands remain relaxed on the upper jacket. From 00:05.500 to 00:08.000, keep the separated affectionate pose nearly still, with two readable profiles and both identities clearly recognizable. Let hair strands settle and candle highlights flicker faintly. Preserve the square chest-up framing, high collars, black lace bow, earrings and carved-wood room through the last frame. No text, watermark, extra person, costume transformation, nudity, tongue or open mouth. overall_soundscape: Newly generated quiet interior room tone, faint distant candle hiss and one soft fabric rustle during the lean and return. No dialogue, vocalization, kissing sound effect, exaggerated breathing or copied source audio. The room sound remains continuous across all eight seconds, without cuts or loud impacts. non_diegetic_music: A newly composed, very quiet romantic dark-fantasy instrumental: soft sustained cello and a few sparse felt-piano notes, slow and restrained, beneath the room tone. Ease gently under the final separated pose, without a dramatic swell. No vocals, recognizable borrowed melody or heavy beat.

曝光 15 · 评论 0 · 点赞 1 · 书签 0