巨物感留住了，“扶绳走一步”却没交代清楚。 把 @LiuLoky50572 的吊桥静态图交给本地 H3，做了 8 秒单镜头，只跑一版。巨拱、雾谷和远处红眼都… · @superfang119 · TwKit

@superfang119

巨物感留住了，“扶绳走一步”却没交代清楚。 把 @LiuLoky50572 的吊桥静态图交给本地 H3，做了 8 秒单镜头，只跑一版。巨拱、雾谷和远处红眼都还在，披风的变化也给画面加了风感。 我要求的是手不松绳、只走一步、两只靴子保持可见。实际抽帧里，中段手、绳索与武器交叠，持续握绳看不清；动作中后脚又被披风遮住，完整的抬脚—落脚过程没法确认。不能把画面好看当成动作已经通过。 我会留下它的雾谷氛围。要展示清楚的一步行走，先修手部接触和后脚可见性。音轨未主观试听，桥板承重与全程绳索连接也没做逐帧验证。 原帖有三张不同场景图，这轮只用了第1张吊桥图。动作和声音是我另行设计的，下面是实际提交的完整 H3 提示词与参数。 Source / image reference: @LiuLoky50572 https://x.com/LiuLoky50572/status/2111156014078886017 The source post contains three still images. Only photo1, the bridge composition, was supplied to this H3 run, as an unchanged 800x1200 medium-image download. The other scenes were not model inputs. Ref2VA is an image reference, not an exact first-frame or 3D geometry lock. The full H3 prompt is newly written from the visible image, not the source author's published execution prompt. All motion and sound are newly designed. Local ComfyUI / H3 Ref2VA; one generated 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 copy of the native H.264/AAC file, with no 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: 211115601407888. AI-generated H3 video. The prompt below is copied from the actual submitted workflow. Full actual H3 prompt: subject_definitions: <Subject 1> is the adult female warrior in <Picture 1>, with long ash-brown hair, weathered dark leather armor, wrapped gloves, a skull-shaped shoulder pauldron, a layered ragged sand-colored skirt and cape, and tall strapped armored boots. Keep her recognizable appearance, proportions and costume. <Subject 2> is the narrow damaged rope-and-timber suspension bridge in <Picture 1>. Preserve the large near-left wooden post, thick handrail rope running diagonally across the foreground, hanging vertical ropes, irregular planks beneath the warrior and the bridge receding toward frame right into mist. <Subject 3> is the distant misty canyon with enormous rib-like stone arches, tall eroded rock columns and a gigantic head-shaped figure with one small red eye glow in the upper background of <Picture 1>. Keep these landmarks in the same approximate locations and at a vastly larger scale than the warrior. <Subject 4> is the warrior's single long pole weapon with a narrow pointed blade extending toward frame right in <Picture 1>. Keep it intact and separate from the bridge rope. The near-left hand holds the handrail, while the other hand supports the weapon. <Picture 1> is the composition reference for the entire single shot: a full-body warrior on a broken bridge above a fog-filled abyss, with the rough foreground post and rope framing her and enormous stone arches and a red-eyed distant giant above. It is an image reference, not a pixel-exact first-frame lock. summary: [reference generation] Generate an 8-second vertical dark-fantasy film using one static image reference. In one continuous restrained shot, the adult warrior cautiously advances one step on the suspension bridge and settles, maintaining her grip on the handrail and her weapon. Wind moves her torn cape and hair; low mist drifts through the huge stone arches. Retain the source's desaturated bone-white, gray-brown and charcoal palette, rough weathered materials and immense scale contrast. All motion and sound are newly designed; no video or audio reference is provided. retention_analysis: <Subject 1> (00:00.000-00:08.000): fully_preserved - retain the adult warrior, long hair, skull pauldron, leather armor, ragged cape and tall boots while adding only a cautious single step and subtle head movement. <Subject 2> (00:00.000-00:08.000): fully_preserved - preserve the bridge path, handrail and hanging rope connections, nearby post and plank arrangement. Allow only a small local sag and sway under the step, without replacing the bridge geometry. <Subject 3> (00:00.000-00:08.000): fully_preserved - keep the giant stone arches, background head and one dim red eye in their approximate positions. Only mist and subtle light attenuation move; no giant attack or new creature appears. <Subject 4> (00:00.000-00:08.000): fully_preserved - keep one solid pole weapon, its grip and blade orientation, distinct from the rope; no extra weapon or object transfer. <Picture 1> (single-shot framing): weak_reference - preserve the full-body composition, near-left framing post, rightward bridge depth and giant background scale without requiring identical pixels. The full boots and their contact with planks remain visible. detailed_description: [Shot 1] One continuous 8-second vertical 2:3 composition following <Picture 1>. The adult warrior stands in the lower middle of the image, facing mostly frame right with her head angled downward toward the next plank. One gloved hand grips the coarse foreground handrail while the other supports the single pole weapon, blade pointing toward frame right. Her full figure and both boots are visible. A large blurred timber post occupies the near-left edge. The damaged bridge planks and connected vertical rope hangers recede diagonally into the fog at frame right. Enormous curved stone ribs, eroded pillars and the distant giant's small red eye remain visible above her. Soft diffuse overcast light and gray-beige mist separate several depth layers. Keep a gritty cinematic fantasy illustration look, with rough timber, frayed rope, worn leather and stone surfaces. From 00:00.000 to 00:01.500, hold the initial full-body bridge composition. Wind draws a few long hair strands and loose cape strips toward frame left; thin mist drifts between the arches. The distant red eye stays small and dim. Begin an extremely slow camera push, without orbiting or changing which side of the rope the viewer sees. From 00:01.500 to 00:04.500, the warrior cautiously transfers her weight and takes one short step toward frame right along the existing planks. The forward boot lifts just clear of the boards and plants on the next visible plank; the trailing boot remains supported before it draws in slightly. Keep the hand on the same continuous rope, sliding a short distance without letting go. Preserve clear glove-rope contact. The weapon moves with her torso but remains a single solid pole, not fused into the handrail. The local plank dips a little and the nearby hanging ropes tension gently; bridge motion is small rather than a violent bounce. A faint wood creak, rope fiber strain and one restrained boot contact accompany the weight transfer. From 00:04.500 to 00:06.500, she finishes the single step and stops. Both boots are firmly supported. She lifts her chin slightly toward the far bridge while maintaining her grip and controlled posture. Cape strips settle into a small wind-driven flutter. The camera continues its tiny forward drift, keeping the full boots visible and the near-left post, handrail and giant stone ribs in a consistent arrangement. Do not add another step or move her off the bridge. From 00:06.500 to 00:08.000, hold the settled pose and nearly stop the camera. Let the mist open briefly around the far head silhouette and dim red eye, then soften again. The giant remains still. Preserve the narrow bridge path into frame right, intact rope connections, visible boot support and the same weapon through the final frame. No cuts, costume transformation, extra person, new creature, text, logo, gore or combat impact. overall_soundscape: Create a newly generated quiet canyon wind with distant low airy rumble, sparse cape and leather movement, subtle rope strain, a small timber creak and one restrained boot contact during the step. Keep the giant silent and distant. No dialogue, vocalization, copied source audio, breaking bridge or abrupt explosion. non_diegetic_music: Add a newly generated very soft low-string drone with a sparse dark ambient pulse, leaving space for the wood and rope sounds. Hold tension without a heroic climax and ease slightly under the final still pose. No vocals, recognizable borrowed melody or heavy beat.

曝光 15 · 评论 0 · 点赞 0 · 书签 0