这轮我会留开头全景。小兵、左侧装甲车和远山一摆，机甲的巨物感就有了。 H3也做出了巨脚接地后的扬尘，不过头盔近景换了朝向，回全景早于我写的11.75秒。做这… · @superfang119 · TwKit

@superfang119

这轮我会留开头全景。小兵、左侧装甲车和远山一摆，机甲的巨物感就有了。 H3也做出了巨脚接地后的扬尘，不过头盔近景换了朝向，回全景早于我写的11.75秒。做这类科幻预告，我更喜欢大小主体之间的尺度反差；装甲再细，也得让观众看得出它有多大。 参考 @EiiKwesiKay 的 Megagiant，用4张原片抽帧，本地H3一次生成15秒，没有重抽。原帖未公开提示词，下面这份是我根据画面重新写的。 按首、中、尾和动作节点抽帧复核；生成音轨已完整解码，未做主观听审。以下附完整实际提示词和参数。 Local ComfyUI / H3 Ref2VA, one generated sample, no reroll. Four image references extracted from the source at 0 / 5.5 / 7.5 / 10 seconds: wide scene, soldier helmet, machine lower body, foot. These are image references, not a first-frame lock; the source video/audio were not model inputs. 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, CFG 1; lowres_scale 0.6, transition_step 6, rho 0, w_min 0.5 / w_max 1. Upscaler: h3_upscaler_lms_v0.1.safetensors. Attention: comfy kitchen attention + sol-attn (tau 1.3, start 0.2 / end 1); sigma shift video 12 / audio 3. Negative conditioning: original ConditioningZeroOut. Canvas: 1:1 (Square), 1.5MP, multiple 32 -> 1248x1248, 24fps. Requested 15s, native 362 frames / 15.083333s. Delivery uses the first 360 frames and first 15s of generated audio, H.264 CRF19 / AAC stereo 32000Hz, exactly 15.000s; no speed change or interpolation. Video VAE: minimax_h3_video_vae_int8_convrot.safetensors Audio VAE: minimax_h3_audio_vae_fp32.safetensors Seed: 211145724083826. The full prompt below is copied from the actual submitted workflow. Source / visual reference: @EiiKwesiKay https://x.com/EiiKwesiKay/status/2111457240838262850 AI-generated H3 recreation; this is a newly written prompt, not the source author's published prompt. Full actual prompt: subject_definitions: <Subject 1> is the colossal humanoid machine in <Picture 1> and <Picture 3>: weathered ivory and gunmetal armor, bronze seams, a tall narrow vertical head crest, broad angular shoulders, hanging cables, long articulated arms, heavy segmented legs, and a bright circular white chest core. It towers far above the distant hills. <Subject 2> is the solitary adult armored soldier in <Picture 1> and <Picture 2>: dark bronze and black worn combat armor, a rounded protective helmet with a dark transparent visor, a visible human face behind the visor, and a long equipment or weapon silhouette secured vertically beside the back. Keep the helmet and armor identity consistent. <Subject 3> is the golden grassland and low dusty hills in <Picture 1>, under a pale warm sky; a tiny low armored vehicle with a small flag stands to the far left. These familiar small objects establish the machine's extraordinary scale. <Picture 1> is the square wide composition reference for [Shot 1] and [Shot 6], with <Subject 2> small at bottom center, <Subject 1> filling the upper background, and the vehicle left of the soldier. <Picture 2> is the soldier's helmet profile reference for [Shot 3]. <Picture 3> is the low-angle machine lower-body composition reference for [Shot 4]. <Picture 4> is the ground-level heavy foot reference for [Shot 5]. summary: [reference generation] Create a 15-second square cinematic science-fiction sequence: a lone soldier faces a mountain-sized machine across an autumn grassland. Move from a scale-establishing wide shot through an over-the-shoulder view and helmet close-up, reveal the machine's heavy movement, show one foot planting into the grass with dust, then return to the opening wide composition. Use the four pictures as appearance and composition references, not as video or audio inputs. Generate a new soundtrack. retention_analysis: <Subject 1> (appears in [Shot 1], [Shot 2], [Shot 4], [Shot 5], [Shot 6]): fully_preserved - retain the ivory segmented armor, bronze seams, enormous scale, articulated limbs, hanging cables, tall crest, and circular chest core where visible. Its silhouette remains the same machine across cuts. <Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 6]): fully_preserved - keep the worn bronze helmet, visor, armor, and back-mounted equipment; the wide-shot figure and close-up are the same adult soldier. <Subject 3> (appears throughout): fully_preserved - maintain the golden grass, warm dusty hills, pale sky, and wide-shot left-side vehicle as spatial scale anchors. Close shots may exclude objects outside their field of view. <Picture 1> ([Shot 1] and [Shot 6] composition): weak_reference - follow the central soldier, left vehicle, distant ridge, and towering machine arrangement without requiring pixel-identical first or last frames. <Picture 2> ([Shot 3] composition): weak_reference - preserve the close side profile, visor material, and warm edge light. <Picture 3> ([Shot 4] composition): weak_reference - preserve the monumental lower-body perspective and articulated plates. <Picture 4> ([Shot 5] composition): weak_reference - preserve the foot, grass scale, and dusty ground-contact view while animating a complete step. detailed_description: Use cinematic realistic science-fiction imagery, warm desaturated ochre grass, dusty beige atmosphere, textured metal and soft directional daylight. The entire sequence is 15 seconds, square 1:1, with six shots joined by five deliberate hard cuts and no captions, logos, or subtitles. [Shot 1] Establish a very wide eye-level view based on <Picture 1>. <Subject 2>, the solitary bronze-armored soldier, stands still at bottom center, seen from behind with both feet planted and back-mounted equipment visible. <Subject 3> stretches toward low dusty hills; the tiny flagged vehicle remains at far left. Beyond the ridge, <Subject 1> occupies most of the upper frame, its immense shoulders and head crest extending beyond the top boundary, circular white core facing the camera. The machine slowly advances, with subtle armor and cable movement; fine haze softens its lower legs. A few tiny dark flying silhouettes pass near its left arm. Keep the soldier small and stable so the size comparison remains legible. Low wind and a distant mechanical rumble begin. [Shot 2] At 00:02.750, cut to an over-the-shoulder view. The soldier's bronze shoulder, helmet edge and back equipment occupy the right foreground; the huge machine's arm and legs dominate the background above the hills. Use a restrained forward camera creep, not a spin. The soldier makes only a slight upward head movement. Grass stirs; the mechanical rumble grows without explosions. [Shot 3] At 00:04.750, cut to a tight side-profile close-up guided by <Picture 2>. Show the same scratched bronze helmet, dark transparent visor and human eye and cheek behind it. The adult soldier looks upward toward the off-screen machine, with a restrained breath and a tiny head tilt. Warm sky light edges the helmet. Keep the visor solid and the helmet geometry intact; do not replace the face with a robotic mask. The camera holds the profile with a very small push-in. [Shot 4] At 00:06.750, cut to a low-angle medium-wide view guided by <Picture 3>. The machine's waist, long hanging hands and immense segmented legs fill the frame; the head is outside the crop. One leg slowly shifts forward, revealing dark joints beneath ivory armor; cables sway with the movement. Maintain the same grassland and dusty horizon. Heavy servo strain and low metal creaks accompany this mass, without rapid combat action. [Shot 5] At 00:08.750, cut to a ground-level close-up guided by <Picture 4>. A massive armored foot descends from above frame toward golden grass with scattered reddish wildflowers. Its broad sole makes one clearly readable contact around 00:10.000, compressing the grass and pushing a low dust cloud outward. Hold long enough to see the foot settle and dust drift. Do not turn the foot into a vehicle, add extra toes, or switch feet mid-contact. A single deep impact and brief rattling metal follow the visible contact. [Shot 6] At 00:11.750, cut back to the original wide arrangement from <Picture 1>. The soldier remains bottom center, the flagged vehicle stays far left, and the same giant faces forward beyond the ridge. Keep the chest core bright, the shoulders enormous and the soldier dwarfed by its scale. Dust lingers at ground level; the machine finishes its weight shift and becomes nearly still. Hold this composition to the final frame at 00:15.000, with no battle, weapon fire, or sudden new subject. overall_soundscape: Generate a new continuous dry-grassland wind ambience with sparse grass rustle, a quiet close breath in [Shot 3], distant machine rumble, slow servo and cable creaks, and one deep footfall aligned to the visible ground contact in [Shot 5]. The sound perspective follows each shot. No dialogue, crowd, gunfire, explosions, or reused source audio. non_diegetic_music: A newly generated restrained low synth drone with slow sub-bass pulses and sparse metallic harmonic texture supports the sense of enormous scale. Keep it below the mechanical effects, build slightly toward the footfall, then decay into the final held wide shot. No vocals, recognizable tune, or triumphant orchestral swell.

曝光 78 · 评论 0 · 点赞 1 · 书签 0