Claude Fable 5.2 prompt hands-on test: MiniMax H3-Image image generation
The author used nine portrait prompt rules summarized by Claude Fable 5.2 to run a hands-on test of H3-Image, released by MiniMax the same day, generating a nine-grid set of at-home portraits, and reported that details such as strands of hair, the faint sheen on the collarbone, and the ribbing of cotton fabric were clear, with the face remaining largely consistent across different poses. H3-Image is derived from H3 and supports text-to-image, image reference, and image editing; it can generate 4K images and produces an image in about 10 seconds. The author speculates that its approach is to distill the video foundation model after removing audio, thereby transferring the video model's understanding of three-dimensional lighting to static portraits.
View original postHoly crap, Fable 5 is truly outrageous, seriously incredible, the prompt it wrote can actually make Grok generate results comparable to seedance 2.5 in effect and texture, at 6 times lower cost! Prompt: Main character: a young Korean woman, around twenty-five years old, with delicate natural everyday makeup, wearing a wide-brimmed beige straw hat (the brim has wide dark brown stripes), a light green off-shoulder cross-ruched dress, pearl earrings and a thin gold bracelet, with long dark brown hair falling naturally or lightly pinned up under the straw hat, a warm and approachable personality. Maintain consistent identity, clothing, hairstyle, and appearance throughout the video. Realistic skin texture, light makeup. Location: a bright afternoon, real spectator stands at the Eastbourne tennis tournament. The green grass court is in the foreground, wooden and plastic seats, with other spectators in light suits and summer casual wear in the background. Strong natural sunlight shines from above, with occasional cloud movement bringing changes in light, shadow, and exposure, a warm and relaxed sports event atmosphere. Focus always remains on her natural reactions and personal moments. Visual style: hyper-realistic documentary authenticity. Real improvised behavior. Natural body language. An unscripted sense of everyday-life footage. Strong environmental authenticity. Rich real-world detail and believable human movement. Camera style: the aesthetic of early-2000s consumer DV camcorders. A friend casually recording everyday-life moments. Strong handheld shake, imperfect framing, frequent autofocus hunting, lens breathing, exposure fluctuations when moving between sunlight and shadow, occasional motion blur, slight rolling shutter, moderate digital compression artifacts, faded colors, soft contrast, slight sensor noise. No stabilization. No cinematic camera movement. No modern color grading. 00:00–00:02 She is on the green seats in the spectator stands, lightly holding the brim of her straw hat with her right hand, smiling as she looks toward the court ahead. A breeze moves the edge of the brim and strands of hair. She smiles naturally while the camera struggles to keep focus on her face, with obvious handheld shake. 00:02–00:04 The camera follows her from the side as she turns slightly to watch the match, notices an exciting moment, and her expression becomes lively. The framing drifts off-center as the operator tries to keep up with her reactions. Autofocus repeatedly switches between her face and the distant court. 00:04–00:06 A close capture of her warm smile as she seems amused by a good shot on the court, her shoulders trembling slightly. Sunlight through the straw hat casts moving soft shadows on her face. The shot has natural breathing and slight exposure fluctuations. 00:06–00:08 A slightly wider framing, she sits comfortably in her seat, body relaxed, left hand on her lap, continuing to watch the match. Occasionally she tucks hair behind her ear or adjusts the hem of her dress. The handheld shot has natural drift, and the light changes as clouds move. 00:08–00:10 A close-up profile. She notices the camera (her friend), turns around, gives a sincere warm smile, and lightly raises a hand in greeting or adjusts her hat. The camera catches this moment a little late, then the recording ends naturally. Audio: natural ambient sound only—gentle wind, the crisp sound of tennis rackets hitting the ball in the distance, faint murmurs and occasional applause from the stands, slight creaking of seats, the rustle of grass or surrounding flags. Subtle event atmosphere. No music. No sound design. No voiceover. Goal: capture a warm slice of life of a real tennis spectator, as if it were a forgotten early-2000s home video—improvised, imperfect, authentic, warm, and highly convincing.
View original
卧槽,Fable 5真的逆天啊,真的太牛逼了, 它写的提示词竟然能让Grok生成堪比seedance 2.5效果和质感,成本低6倍! Prompt: 主要角色:年轻韩国女性,二十五岁左右,精致的自然日常妆容,戴着宽檐米色草帽(帽檐有深棕色宽条纹),穿着浅绿色露肩交叉褶皱连衣裙,戴珍珠耳环和细金手链,深棕色长发在草帽下自然垂落或轻盘,温暖而亲切的个性。在整个视频中保持一致的身份、服装、发型和外貌。逼真的皮肤纹理,淡妆。 地点:明媚的午后时分,真实的东bourne网球锦标赛观众席。绿色的草地球场在前景,木质与塑料座椅,背景中有其他穿着浅色西装和夏日休闲服装的观众。强烈的自然阳光从上方照射,偶尔云层移动带来光影和曝光变化,温暖而轻松的体育赛事氛围。焦点始终在她的自然反应与个人时刻上。 视觉风格:超现实主义纪录片真实感。真实的即兴行为。自然的肢体语言。无剧本的日常生活片段感。强烈的环境真实性。丰富的现实世界细节和可信的人类动作。 摄像风格:2000年代初消费级DV摄像机的美学。朋友随意记录日常生活瞬间。强烈的手持抖动,不完美的构图,频繁的自动对焦搜索,镜头呼吸,在阳光和阴影间移动时的曝光波动,偶尔的运动模糊,轻微的滚动快门,中等数字压缩伪影,褪色的色彩,柔和的对比度,轻微的传感器噪点。没有稳定。没有电影化的摄像机移动。没有现代色彩分级。 00:00–00:02 她在观众席绿色座椅上,右手轻扶草帽帽檐,微笑地看着前方球场方向。一阵微风吹动帽檐边缘和发丝。她自然地微笑,而摄像机努力保持焦点在她的脸上,手持明显抖动。 00:02–00:04 摄像机从侧面跟随她,她微微侧身观看比赛,注意到精彩瞬间,表情生动。构图偏离中心,因为操作者试图跟上她的反应。自动对焦反复在她的脸部和远处的球场之间切换。 00:04–00:06 近距离捕捉她温暖的笑容,她似乎被场上的一个好球逗乐,肩膀轻颤。阳光透过草帽在她脸上投下移动的柔和阴影。镜头有自然的呼吸和轻微曝光波动。 00:06–00:08 稍宽的构图,她舒适地坐在座位上,身体放松,左手放在腿上,继续观察比赛。偶尔用手拨弄耳边头发或调整裙摆。手持镜头带有自然的漂移,云层移动时光线发生变化。 00:08–00:10 侧脸近距离轮廓。她注意到摄像机(朋友),转过身来,露出真诚温暖的微笑,轻轻举手致意或调整帽子。摄像机稍晚捕捉到这一刻,然后自然结束录制。 音频:仅自然环境音——轻柔的风声、远处网球拍击球的清脆声、观众席微弱的低语和偶尔掌声、座椅轻微的吱嘎声、草地或周围旗帜的沙沙声。细微的赛事氛围。没有音乐。没有音效设计。没有旁白。 目标:捕捉真实的网球赛事观众的温暖生活片段,仿佛一段被遗忘的2000年代初家庭录像——即兴、不完美、真实、温暖且极具说服力。
Body · AI translation
Holy crap, everyone stand up! Claude Fable 5.2 has completely ascended to godhood—distilling a video foundation model to paint still portraits is just a one-sided massacre of the laws of physics, holy crap!! The Nine Secret Formulas for AI-generated pure-desire portraits summarized by Fable 5.2, plus the exclusive top-tier prompts it helped me optimize, are being shared open-source with everyone, no holds barred!! I can't even help but drop a hot take, brothers: in AIGC, anyone still trial-and-erroring image prompts line by line is the most miserable wage slave of the compute era!!
This afternoon I fed the image-generation secret formulas summarized by Claude Fable 5.2 into MiniMax's newly released H3-Image, and the nine-grid set of at-home portraits it actually produced in testing left me stunned—seriously incredible, it feels like my youth has come back, ahhh!
Before, the scariest things about running portraits were the inflatable-doll fake-skin feel and stiff poses, but this time, running according to the aesthetic skeleton Fable 5.2 broke down, the realistic texture straight-up leaves traditional models several tiers behind,
Rule one: define the persona with a specific age plus recognizable features—write an East Asian adult woman around 25 years old, then lock down a single tear mole, brow shape, and hair length,
Rule two: describe the figure head-on only once, and leave the rest to cut and fabric—for example, a fitted white ribbed cotton thin-strap camisole, round neckline, hem tucked above the waistline,
Rule three: use half-finished actions to create a candid feel—stretching, brushing hair, tying a ponytail—and in the nine-grid, keep only two frames looking at the camera,
Rule four: define the tone with light position and exposure; don't write a single word about focal length or aperture—only write diffused morning light falling on the face through a white gauze curtain, exposure based on the face,
Rule five: use close-ups to convey bone structure and skin; reserve one frame solely for the collarbone and shoulder strap, specifying visible tiny pores and fine vellus hair, no skin smoothing,
Rule six: use a white scene to bring out an airy feel—white walls, white bedsheets, white gauze curtains, with white occupying more than half the frame,
Rule seven: break the film feel into separate parts—high-key slightly overexposed, low contrast, low saturation, fine grain; don't use a single vague film look,
Rule eight: at the end, use an avoidance list to clamp down on AI plastic-ness—specify plastic skin, excessive skin smoothing, doll face, darkened face, graying whites,
This writing approach happened to coincide with @MiniMax_AI's official release today of H3-Image, an image generation model derived from H3, featuring text-to-image, image reference, and image editing capabilities, 4K image generation, and it gets the job done in just 10 seconds!!
One look at the attached images shows how absurd it is—stray strands of hair, the faint sheen on the collarbone, the ribbing of the cotton fabric, and that little mole under the eye are all crystal clear, and the face stays basically consistent across different poses,
My understanding is that it stripped the audio from its own mature video foundation and distilled it; video models are constantly understanding the 3D continuity of light and shadow in the real world every day, so turning back to paint a still portrait is just dimensional-reduction harvesting,
A top-tier large model translates the cinematographer's aesthetics into plain human language, the video foundation delivers an image in ten seconds, and the full set of prompts is already in the comments—brothers, help yourselves and go run them. #H3Image