登録して招待リンクを共有すると、動画再生報酬と紹介報酬を獲得できます。

検索結果 consistency
consistency コミュニティ
1つのキーワードが1つのコミュニティです。
コミュニティ作成
アカウント
見つかりません
consistency を含む検索結果
📚 ReActもRAGもTree of Thoughtsも、論文ごとにバラバラだったエージェント設計を、同じAPIで動かして比較できたら最高だと思いませんか?それを実現した「35パターン全部入り」のリポジトリです。 タイトル: FareedKhan-dev/all-agentic-architectures URL: 📦 概要 本リポジトリは、プロダクション品質のエージェントAIパターンを35種類実装したPythonライブラリ兼「生きた教科書」です。すべてのアーキテクチャが同じ.run(task)メソッドを持ち、同一形式の結果を返すため、下流のコードを変えずにパターンを差し替えられます。 ❓ 解決する課題 エージェントの設計パターンは論文ごとに散らばっていて、実装も様式もバラバラでした。これを統一インターフェースの下に集約し、横並びで試せるようにしたのが最大の価値です。 💡 中核の工夫と提案手法 中心にあるのが「決定論的ピッカーの規律」です。 ・LLMのスコアリングに丸投げせず、まずLLMに真偽値や列挙型などカテゴリ的な特徴をコミットさせる ・最終判断はPythonのロジックで合成する これにより、スコアが平坦に潰れる「LLM-as-Scorer」の病理を緩和します。35アーキテクチャ中13で採用されています。 🎯 カバー範囲とユースケース 推論・内省(Reflection、Self-Discover)、探索(Tree of Thoughts、LATS)、RAG(Corrective/Self/Adaptive/GraphRAG)、メモリ(MemGPT、Voyager)、ツール・行動(ReAct、SWE-Agent)、マルチエージェント(Debate、STORM)など8系統を網羅。各パターンに実行済みのJupyterノートブックが付き、本物のLLM出力に基づく再現可能なリファレンスになっています。 📊 注目ポイント ・コアはLangGraph。NebiusやOpenAI、Anthropic、Ollamaなど主要プロバイダーに対応し、切り替えは環境変数1つ ・pytestで283件のテストがパス ・17タスクのベンチマークで直近42問中33問正解(成功率78%)。ReflectionやSelf-Consistencyが好成績でした #AIエージェント# #LangGraph#
もっと見る
AI画像生成、ついにここまで来たか。 新モデル「MiniMax H3-Image」で、実写映画のキャラクターシートを制作。 正面・側面・背面の3アングルを1枚で生成。 顔の造形はもちろん、衣装の構造、アクセサリー、素材の質感まで質感がかなりいい。 映画やゲームのキャラクターデザインにも活用できそう。 プロンプト: Create an ultra-photorealistic live-action character turnaround sheet for an original Cleopatra-inspired fantasy queen. Aspect ratio 16:9, 4K resolution. CHARACTER: An exceptionally beautiful 23-year-old Japanese woman with the fresh, charming appearance of a contemporary Japanese actress. She has a youthful, naturally cute face with large expressive dark brown eyes, softly defined double eyelids, a small refined nose, softly rounded cheeks, and naturally full pink lips. Her facial features are modern, delicate, and naturally balanced, with an approachable yet captivating beauty. Realistic skin texture, visible pores, subtle imperfections, and natural facial details. BODY PROPORTIONS: A slim, graceful adult woman with a distinctly slender torso, narrow ribcage, flat stomach, and beautifully defined waistline. She has a naturally large, full bust that contrasts elegantly with her slim upper body and narrow waist. Her abdomen remains flat and slender, with subtle natural muscle definition. Slim arms, graceful shoulders, slender hips, and long, shapely legs. Maintain a delicate, slim overall silhouette despite her prominent bust. Realistic human anatomy and believable body proportions. HAIRSTYLE: Shoulder-length, sleek, straight jet-black bob. The hair falls naturally to her shoulders with a smooth silky texture and slightly inward-curving ends. Soft center part. No long hair, braids, ponytails, or elaborate updos. FACIAL MAKEUP: Elegant black Cleopatra-inspired eyeliner extending subtly beyond the outer corners of her large eyes. Distinctive ceremonial vermilion-red markings beneath both eyes. Two thin crimson lines extend downward from each lower eyelid toward the upper cheeks, accompanied by several tiny symmetrical red dots. The markings are delicate, precise, and mysterious. Soft natural pink lips and refined contemporary makeup. COSTUME: An exquisite Cleopatra-inspired ceremonial costume created for a premium live-action fantasy film. The costume is luxurious, lightweight, elegant, and feminine. UPPER GARMENT: A beautifully tailored ivory-white silk crop top made from thin, lightweight, opaque fabric. A moderately low, gently open neckline reveals tasteful cleavage. The garment is carefully constructed to fit her naturally full bust while keeping her torso and waist visually slender. Delicate gold-trimmed shoulder straps and subtle ornamental gold chains. Fine gold embroidery along the edges. Her midriff and navel are completely visible. The fabric follows her natural figure without excessive compression or artificial shaping. LOWER GARMENT: A flowing ivory-white silk skirt worn low on her hips, leaving her slender waist, flat abdomen, and navel exposed. The skirt features a dramatic, high-cut slit along her right leg. The slit extends from the upper thigh down to the hem, clearly revealing her right thigh, knee, and lower leg. The exposed leg is fully visible in the front view. The remaining skirt fabric falls in elegant, lightweight folds around her other leg. The design creates a graceful asymmetrical silhouette, balancing flowing fabric with the exposed leg. The slit must look like an intentional, beautifully tailored opening in the skirt rather than torn fabric. A refined ornamental gold belt rests around her hips, decorated with intricate Egyptian-inspired engravings and delicate hanging gold chains. The fabric is thin, airy, and softly draped, but completely opaque. NECKLACE: A delicate, elegant gold necklace featuring a small polished turquoise gemstone. The turquoise stone is modest in size, approximately 1.5 centimeters wide. It rests at the center of her upper chest, just above the neckline. The gemstone is surrounded by a thin, intricate gold setting. The necklace should feel refined and understated rather than oversized or ceremonial. The turquoise remains clearly visible but does not dominate the costume. GOLD JEWELRY: Multiple beautifully crafted gold bracelets on both wrists. Elegant engraved gold upper-arm cuffs. Delicate gold rings. Long, refined gold earrings. All jewelry is handcrafted with realistic metallic reflections and intricate detailing. HAIR ACCESSORIES: A delicate gold circlet resting across her forehead. A small turquoise accent at the center. Fine gold hair ornaments near her temples. The accessories must preserve the clean silhouette of her straight black bob. FOOTWEAR: Elegant ancient Egyptian-inspired sandals with thin gold leather straps. COLOR PALETTE: Ivory-white silk. Warm antique gold. Small vivid turquoise accents. Jet-black hair. Natural luminous skin. Deep vermilion ceremonial markings. CHARACTER SHEET COMPOSITION: Create a professional live-action film costume reference sheet. Show exactly THREE full-body views of the SAME woman arranged horizontally. LEFT: Full-body FRONT VIEW. Facing directly toward the camera. Neutral standing pose. Arms relaxed slightly away from her body. Clearly show her beautiful face, full bust, narrow waist, exposed navel, small turquoise necklace, and the high slit revealing her right thigh and knee. CENTER: Full-body SIDE PROFILE VIEW. Facing exactly 90 degrees to the right. Clearly show her straight bob silhouette, slim torso, natural bust profile, exposed midriff, and layered skirt construction. RIGHT: Full-body BACK VIEW. Facing directly away from the camera. Clearly reveal the back of her bob haircut, costume straps, gold accessories, narrow waist, ornamental belt, and flowing skirt. CONSISTENCY: All three views must depict exactly the same woman. Identical facial identity. Identical body proportions. Identical slim torso and full bust. Identical shoulder-length black bob. Identical costume design. Identical small turquoise necklace. Identical gold jewelry. Identical skirt slit placement. Identical fabric materials and colors. All three figures must appear at exactly the same scale and height. CAMERA: Professional full-body studio fashion photography. Eye-level camera. 85mm portrait lens. Minimal perspective distortion. Consistent framing across all three views. LIGHTING: Soft, natural studio lighting. Realistic skin tones. Subtle shadows. Accurate gold reflections. Natural fabric textures. BACKGROUND: Clean neutral warm-gray studio background. Subtle floor shadows. No scenery. VISUAL STYLE: Ultra-photorealistic live-action cinema. Real Japanese actress. Natural skin pores and fine facial details. Authentic human anatomy. Realistic silk fabric and handcrafted jewelry. Premium Hollywood fantasy costume design. Professional film wardrobe photography. Extremely detailed 4K quality. The final result must look like three real studio photographs of the same beautiful young Japanese actress wearing an elaborate Cleopatra-inspired royal costume. NEGATIVE CONSTRAINTS: No CGI. No 3D rendering. No anime. No illustration. No plastic skin. No exaggerated or distorted anatomy. No thick torso. No wide waist. No bulky body. No oversized turquoise necklace. No transparent clothing. No long hair. No braids. No text. No labels. No logos. No watermarks. No additional characters. No cropped feet. @MiniMax_AI #H3Image#
もっと見る
商品や人物の動画を、ぐるっと360°回しても「同じもの」のまま生成できるでしょうか🔄 OmniShowcase: Omnidirectional Subject-Consistent Video Generation from Multi-View Images 4枚のマルチビュー参照画像から、被写体の一貫性を保った周回動画を生成する手法です。 🧱 注目ポイント1: 3D整合なデータを自動合成 同じ3Dメッシュから、周回ターゲットと90°間隔の4つの参照ビューを描画します。背景は動画インペイントで生成し、各段階をVLMでゲートします。キュレーション費用は合計約120ドルで、約4,000本のShowcase-4Kとして公開予定です。 📐 注目ポイント2: 法線マップを「文脈」として渡す 4枚の参照から3D Gaussianを作り、法線のターンテーブルをWan2.1にLoRAで与えます。ターゲットとのフレーム単位の対応付けが不要で、参照を1枚から4枚に増やすとDINOが0.702から0.799に上がります。 📊 注目ポイント3: 3D指標で大差 ShowcaseBenchで3DreamBoothと比べ、GT-CDは0.1717から0.0824、MEt3Rは0.325から0.190へ改善しました。ユーザースタディでは各基準で70.9〜79.4%の票を獲得しました。 2D参照を増やすより3Dジオメトリが効く、という示唆が得られます。ただし非剛体は苦手です。 #動画生成# #コンピュータビジョン#
もっと見る