这轮我更喜欢丝绸高光，玻璃反而像慢慢移动的雕塑。 一张礼服人像，交给本地 H3 做成 8 秒短片，只跑一版。抽帧里，高领、褶皱和前后玻璃弧都保留了，人物的脸… · @superfang119 · TwKit

@superfang119

这轮我更喜欢丝绸高光，玻璃反而像慢慢移动的雕塑。 一张礼服人像，交给本地 H3 做成 8 秒短片，只跑一版。抽帧里，高领、褶皱和前后玻璃弧都保留了，人物的脸也没被玻璃挡住。 它没完全照单执行：我写约 2 秒眨眼，半秒采样里到约 3.5 秒才看到闭眼；后半段视线转得更靠左。镜头确实推近了，但有没有守住“小于 10%”还没量化。玻璃从画面上跨过衣料，也不能仅凭抽帧说全程没有穿插。 我会把它当服装与材质展示的小样，喜欢这种动作克制的感觉。生成音轨未主观试听，音画同步未核验。 灵感来自 @LiuLoky50572 的静态琉璃人像。这次重新生成了高领长袖、不透丝绸礼服参考图，只借鉴珍珠色、玻璃与柔光主题；没有沿用原图人物或身体构图。下面附 imagegen 与 H3 两段完整实际提示词，以及本轮参数。 Source / visual theme: @LiuLoky50572 https://x.com/LiuLoky50572/status/2111321245153448090 The source is a static image. This is a new, fully clothed fashion adaptation of its pearl/glass/light theme, not an identity or body-composition replica. The source image was not supplied to the successful imagegen call or to H3. Image reference: built-in imagegen, one successful original image, 1086x1448. Model version/seed were not returned. The first source-image edit was refused; its prompt was not used for the successful original image. The successful image prompt is copied below. Local ComfyUI / H3 Ref2VA; one generated sample, no reroll, one image reference: the new ivory high-neck long-sleeve portrait. Ref2VA does not lock the exact first frame. 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 original ConditioningZeroOut. Canvas: 3:4 (Portrait Standard), 1.5MP, original multiple32 -> 1088x1440, 24fps; requested8s, native192frames/8.000000s. Delivery is an unchanged copy of the native H.264/AAC file, no cut, speed change or interpolation. Audio: newly generated AAC 32000Hz, 2 channels. No source audio was provided; the 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: 211132124515344. AI-generated image and H3 video; both actual successful prompts follow. Full actual imagegen prompt: Create one vertical 3:4 high-end fashion editorial image for an animation reference. An unmistakably adult 28-year-old woman in a fully opaque ivory silk high-neck gown with long sleeves, photographed from the chest up in a gentle three-quarter view. Ash-blonde hair in a loose low bun, fine wispy bangs, gray-blue eyes, mature facial proportions, a calm neutral expression. The entire chest and shoulders are covered by the opaque garment. A large clear glass sculpture makes a graceful ribbon-shaped arc behind her and a smaller arc crosses the foreground, both floating apart from her. Optical refraction, rounded glass edges, thin bright specular highlights, distinguish glass from the softly folded silk fabric. Diffuse studio window light, pearl white and champagne palette with faint blush background, soft blurred curtains, refined semi-realistic 3D fashion illustration. Simple elegant composition, face in upper half, sharp hair strands and subtle fabric texture. No text, watermark, extra people or branding. Full actual H3 prompt: subject_definitions: <Subject 1> is the unmistakably adult woman in <Picture 1>, approximately 28 years old, with ash-blonde hair in a loose low bun, fine wispy bangs, gray-blue eyes, mature facial proportions and a calm expression. Keep her recognizable face, hairstyle and natural skin texture. <Subject 2> is her fully opaque ivory silk high-neck, long-sleeve gown from <Picture 1>. Preserve the solid fabric covering the shoulders and entire chest, the gathered collar and the soft diagonal silk folds. Keep the garment intact and opaque throughout. <Subject 3> is the clear sculptural glass ribbon from <Picture 1>: a large rounded arc behind the head and shoulders and a smaller curved foreground arc. Preserve clear refraction, narrow bright edge highlights and separation from the woman and her gown. <Picture 1> is the newly generated fashion portrait composition reference for this entire shot. It supplies the chest-up framing, character, opaque wardrobe, pearl-champagne palette, glass arcs and softly blurred curtain background; it is a visual reference, not an exact first-frame lock. summary: [reference generation] Generate an 8-second vertical 3:4 fashion portrait film from one image reference. In one restrained continuous shot, the adult woman makes a small head turn and a natural blink while transparent glass arcs drift slowly in the foreground and background. Preserve the fully clothed chest-up composition, face, opaque silk folds, clear glass material and soft window lighting. No source video or audio is provided; all motion and sound are newly designed. retention_analysis: <Subject 1> (00:00.000-00:08.000): fully_preserved - keep the mature adult face, ash-blonde loose bun and gray-blue eyes; introduce only a subtle head movement, eye movement and natural blink. <Subject 2> (00:00.000-00:08.000): fully_preserved - keep the high collar, long sleeves, opaque ivory silk and stable fabric coverage, with only slight natural fabric movement. <Subject 3> (00:00.000-00:08.000): partially_preserved - retain the two rounded glass arcs and their optical material while allowing a slow small drift and traveling highlights. The arcs remain separate from the body and never become clothing or intersect the face. <Picture 1> (single-shot composition): weak_reference - preserve the portrait arrangement and material relationships without demanding identical pixels or an exact first frame. Do not extend the composition below the referenced chest-up framing. detailed_description: [Shot 1] One continuous 8-second vertical 3:4 shot. Show only the adult woman's head, covered shoulders and upper torso, following <Picture 1>. Her opaque ivory gown has a high gathered collar and long sleeves. The background is a softly blurred pearl-white and pale blush curtain with diffused daylight entering from frame left. The large transparent glass arc is behind her head and shoulders, and the smaller arc is closer to the camera across the lower foreground. Their bright thin edges and gentle rainbow refraction remain distinct from fabric and skin. At 00:00.000, establish the reference-like calm three-quarter portrait. Hold the composition for the first 00:00.800, allowing a little movement in fine loose hair strands and a narrow highlight traveling along the rear glass arc. From 00:00.800 to 00:03.500, the woman rotates her head only a few degrees toward the lens, keeping the same upright posture and relaxed neutral mouth. She makes one natural blink around 00:02.000. The glass arcs drift a small distance laterally, the rear arc passing behind the shoulder line while the foreground arc moves slowly across the lower edge. Preserve two distinct arcs, rounded clear edges, refraction and empty space between glass and garment. No arc crosses her eyes or mouth. From 00:03.500 to 00:06.000, the camera makes a very small forward drift, less than ten percent tighter, and a slight sideways adjustment. Keep the face clearly visible in the upper half of frame. Silk highlights shift gently across existing folds without changing the gown design or making the fabric transparent. The woman shifts her eyes slightly toward frame upper-left, with no speech or exaggerated expression. From 00:06.000 to 00:08.000, let the glass motion ease and the narrow highlights settle. The woman holds a calm three-quarter gaze, fine hair strands settle and the camera becomes nearly still for the final 00:00.800. Maintain the chest-up crop and all clothing coverage through the final frame. No cuts, costume change, extra person, text, logos or dramatic camera orbit. overall_soundscape: Create a quiet studio ambience with very soft silk movement and a sparse delicate glass-like shimmer as the highlight travels. No source audio is copied. No speech, vocalization, abrupt impacts or exaggerated water splashes. non_diegetic_music: Add a newly generated sparse gentle piano-and-pad instrumental, soft and airy, with a small resolving chord under the final hold. Keep it restrained beneath the glass shimmer. No vocals, recognizable borrowed melody or heavy beat.

曝光 63 · 评论 0 · 点赞 1 · 书签 0