黑白身份没串，裙摆却顶到了画面底边。 参考 @CryY666666 的《双生神女录》，我从五景拼图里选了朱红殿堂的方向。imagegen 另做成年角色图，再… · @superfang119 · TwKit

@superfang119

黑白身份没串，裙摆却顶到了画面底边。 参考 @CryY666666 的《双生神女录》，我从五景拼图里选了朱红殿堂的方向。imagegen 另做成年角色图，再用本地 H3 生成8秒，只跑一版。 黑金在左、象牙金在右，发色和衣色没串。两人最后直视镜头的状态我会留下，红绸也没有挡住脸。 但我要求的是背靠背站定、黑衣先转头、白衣后转头。实际黑衣开场就看向镜头，第一段转头不明显，白衣后面才转回来。后段姿态更正面，黑衣原本遮住的另一只手露出来了；长袍底部也没留足地面余量。要做完整全身的角色展示，这版还不能直接算通过。 参考图本来也没做到四只手全露出，所以 H3 提示词按实际图改成内侧手保持遮挡。这里是新角色、新动作的单场景改编，原拼图没送进模型。只做了抽帧和关键帧细看，转头角度、连续绸带连接和音轨听感未验收。下面是两段完整实际提示词、参数和来源。 Source / theme reference: @CryY666666 https://x.com/CryY666666/status/2111138402649460840 The source is one still collage with five environments, not a video. This sample borrows the black/white adult-woman pairing, Eastern-fantasy styling and vermilion-temple palette to make a new single-scene reference. The source collage was not a model input; identities and costume details are newly created. Head turns, silk movement, sound and music are newly designed, not recovered source motion. The source author's complete execution prompt and model settings were not available in the visible post. Reference image: built-in imagegen, one newly generated1024x1536 image; no source image/video inputs. It is the sole Picture1 supplied to H3. Imagegen requested four visible hands but shows only two outer hands; the H3 prompt therefore keeps inner hands occluded. 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: 2:3 (Portrait Photo), 1.5MP, original multiple32 -> 1024x1536, 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: 211113840264946. AI-generated image and H3 video. The H3 prompt is copied from the actual submitted workflow. Full actual imagegen prompt: Create one vertical 2:3 cinematic photorealistic Eastern-fantasy fashion still, a single uninterrupted scene rather than a collage. Exactly two unmistakably adult fictional women, about twenty-eight years old, posed back-to-back in a deep vermilion ceremonial temple hall. The left woman has long black hair, a small elaborate antique-gold filigree crown, a fully opaque black high-collared silk robe with long embroidered sleeves and a gold sash. The right woman has long silver-white hair, a delicate silver filigree crown, a fully opaque ivory-white high-collared silk robe with long embroidered sleeves and a pale-gold sash. Both chest and shoulders covered, elegant sculpted tailoring, restrained gold ornaments, no transparent body fabric. The black-haired woman faces three-quarter left, the white-haired woman three-quarter right; their shoulder blades lightly meet at center but their heads and forearms remain separated. Both eyes are open, calm confident adult faces, clear individual jawlines. Their arms hang relaxed alongside their own bodies, with two visible natural hands per woman held below the waist; no touching or crossed hands. Three-quarter portrait from crowns to below the knees, with safe empty space above both crowns and a visible band of floor below their hems; two distinct silhouettes remain readable. A single broad translucent red ceremonial silk streamer is tied to a fixed dark-gold column at frame left, behind both women, running across the distant background at waist height and trailing off at frame right; it stays separate from all limbs and costumes. Vermilion lacquer columns, distant red lanterns, dark gold carved beams and warm amber temple light; muted incense haze, deep black and pearl-white clothing against saturated architectural red. Strong editorial composition, detailed silk weave and embroidery, believable skin texture, soft warm rim lighting on both hair colors, shallow depth of field while keeping both faces sharp. No text, watermark, split panels, extra person, extra hand, weapon, nudity or sexualized pose. Full actual H3 prompt: subject_definitions: <Subject 1> is the fictional adult black-haired woman at frame left in <Picture 1>, about twenty-eight years old. Preserve her long black hair, antique-gold filigree crown and earrings, opaque black high-collared robe, embroidered gold birds and flowers, black-and-gold sash, and relaxed visible outer hand. Keep her on the left throughout. <Subject 2> is the fictional adult silver-white-haired woman at frame right in <Picture 1>, about twenty-eight years old. Preserve her silver filigree crown and earrings, long silver-white hair, opaque ivory high-collared robe, gold embroidery and pale-gold sash, and relaxed visible outer hand. Keep her on the right throughout. <Subject 3> is the vermilion ceremonial temple from <Picture 1>: red lacquer columns with gold decoration, dark carved beams, distant warm lanterns, a reflective tiled floor, a shallow layer of incense haze, and one red silk streamer tied around the near-left red column behind both women. <Picture 1> is a newly created vertical single-scene two-person image, guiding adult identities, costume colors, back-to-back positioning, lighting and framing. It is a reference, not a required pixel-identical first frame or a locked 3D set. summary: [reference generation] Make an eight-second uninterrupted portrait shot in <Subject 3>. <Subject 1> and <Subject 2> stand back-to-back without walking or swapping sides. The black-haired woman turns her head a little toward the camera first; the silver-white-haired woman follows after the first settles. A small wave travels through the red silk behind them, then the two women hold their final confident gaze. Preserve the separate black and ivory costumes, two distinct adult faces and the original temple composition. Motion and sound are newly designed; no source collage, source video or source audio is supplied to the model. retention_analysis: <Subject 1> (00:00.000-00:08.000): fully_preserved - keep black hair, gold crown, opaque black robe and left-screen identity. Only a small neck turn, natural blinking, loose hair movement and tiny sleeve response are added. Her visible hand stays relaxed beside her own body. <Subject 2> (00:00.000-00:08.000): fully_preserved - keep white hair, silver crown, opaque ivory robe and right-screen identity. Only her later small neck turn, natural blinking, loose hair movement and tiny sleeve response are added. Her visible hand stays beside her own body. <Subject 3> (00:00.000-00:08.000): fully_preserved - preserve the near-left column, red/gold hall, warm lantern positions, tiled floor and tied red silk. Allow only mild silk movement, flame flicker and haze drift, without replacing the architecture. <Picture 1> (00:00.000-00:08.000): partially_preserved - retain its portrait composition, back-to-back blocking, clothing, identities and palette as image guidance while adding new subtle actions; do not require exact pixel alignment. detailed_description: [Shot 1] (00:00.000-00:08.000), one continuous full-length portrait two-shot, eye level. The women share a central back-to-back line but their heads, forearms and costumes remain distinct. Black-haired woman on the left, white-haired woman on the right. Keep both crowns inside the top edge, both robe hems inside the bottom edge with a narrow floor margin, and both visible outer hands below the waist. Inner hands that are occluded by their bodies do not appear from nowhere. Both chests and shoulders remain fully covered by opaque formal robes. Warm light catches gold embroidery and the red temple remains softly detailed behind them. The single red streamer passes behind the bodies and limbs and is visibly tied to the left column. Camera stays on the same side, with only a very slight slow push that does not crop the crowns or hems; no orbit, cut, full-body spin or new viewpoint. From 00:00.000 to 00:01.500, hold the reference-like back-to-back pose. Each adult face looks slightly outward toward its own side of the lens. Shoulders, hips and feet remain still, with a gentle breath and a few loose hair strands responding to a faint breeze. The silk hangs in its established shallow curve behind them. From 00:01.500 to 00:03.000, <Subject 1> alone turns her head a small amount toward the camera and settles into a composed direct gaze. This is a modest neck movement, not a torso rotation; her gold crown and long black hair stay attached to her head, and her hands remain down. During this interval <Subject 2> holds her initial head direction rather than turning simultaneously. At about 00:02.000 a single small ripple starts near the tied end of the red streamer and travels toward frame right, behind both women, never crossing either face or hand. From 00:03.000 to 00:04.500, <Subject 1> holds her completed gaze while <Subject 2> turns her head a small amount toward the lens and settles. White hair remains white, black hair remains black, and no facial features, crowns, embroidery or robe colors transfer between them. Their backs stay lightly together and bodies do not change sides. The silk ripple continues gently behind their waists and loses energy; no streamer detaches or wraps around an arm. From 00:04.500 to 00:06.000, both women hold their final distinct, calm gaze. A few loose hair strands and hanging ornaments settle. Keep the visible outer hands relaxed on their original sides, without a gesture, additional finger, hand exchange or sudden sleeve expansion. The tied silk eases back toward its reference curve. From 00:06.000 to 00:08.000, keep both characters nearly still in the established pose, with subtle breathing, candlelight flicker and slow haze drift. The camera nearly stops. Maintain both crowns, long robe hems, separate faces, black-left/ivory-right identities, intact column tie and temple landmarks through the final frame. No walking, identity swap, extra person, weapon, costume transformation, nudity, text or watermark. overall_soundscape: Newly generated soft ceremonial-hall room tone, very light silk rustle during the small wave, faint distant lantern flame and spacious but quiet indoor reverberation. No dialogue, vocals, footsteps, impact, copied source audio or sudden sound transition. Keep the eight-second ambience continuous and restrained. non_diegetic_music: A newly composed quiet Eastern-fantasy instrumental with sparse guqin-like plucked strings and a low sustained bowed-string texture. Use a slow, steady mood beneath the silk rustle, a little space between notes and a gentle fade in energy under the final hold. No vocals, recognizable borrowed melody or heavy beat.

曝光 1 · 评论 0 · 点赞 0 · 书签 0