TinyFish · Launch Video Breakdown: Hook, Pacing & Motion Design

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:03

   The Hook

   Product Introduction & Core Value Proposition

   “(No spoken dialogue — Driving, arpeggiated synth bassline with pulsing rhythm, subtle high-frequency synth pads.)”

   On screen

   Web Search and Fetch in OMP powered by TinyFish

   Camera

   Static wide shot, centered text overlay.

   Motion

   Subtle background pixel grid animation, text fade-in.

2. 0:03 – 0:18

   Product Reveal

   Live Terminal Demonstration: Deep Research

   “(No spoken dialogue — Consistent driving synth bassline, atmospheric textures.)”

   On screen

   omp v1.0.4 Welcome back! Tips # for prompt actions # for commands # to run bash # to run python LSP Servers No LSP servers Recent sessions No recent sessions Tip: Tired of typing "keep going"? Just send a '.'. Context reset - 0 messages dropped; session continues. π O GPT-5.A-Sol high tmp 2B what is ox alpha? Do a deep research and get me everything. price, context, limits, who made it, and is it actually glm-5 or what. III. Synthesis 0/2 Assess evidence confidence and unresolved claims Deliver sourced research with practical limits Read skill://use-tinyfish name: use-tinyfish description: The complete web toolkit for your agent. Search the web and get answers in milliseconds. Fetch any URL and get clean fill forms, and extract structured data. Spin up a headless browser for full programmatic control. Use when you need to search bot-protected sites, or automate browser tasks using natural language. TinyFish CLI The complete web toolkit - four tools, one CLI. Start with the lightest tool that can do the job and escalate only when needed. 229 more lines (Ctrl+0: Expand) Output (Resolved path: /Users/pranavjanakiraman/.claude/cache/claude-community/tinyfish/1.0.0/skills/use-tinyfish/SKILL.md) **Pla ! which tinyfish && tinyfish --version || echo T TODO I. Identification 0/2 Identify Ox Alpha product and official listings Trace creator clues and model attribution II. Specifications 0/2 III. Synthesis 0/2 Research Ox Alpha and glm-5 GLM 5.2 vs Ox Alpha: 3B vs 1.05M context. Full pricing, latency and ... March 10: model added glm-5-turbo. March 15: AIHubmix has added edge-Node ... Ox Alpha likely GLM-5.2 - and a GLM 5.5 workhorse? OpenRouter https://www.carcarrouter.ai/blog/ox-alpha-stealth-model-we-know His newest post layers a roadmap guess on top of that read: the GLM-5 update cycle may be short. Ox Alpha is close enough to GLM-5.3 that using ... (Wall: 1.15s | Timeout: 60s) tinyfish search query 'site:openrouter.ai/stealth/ox-alpha context window pricing limits provider modalities' --pretty Output Query: site:openrouter.ai/stealth/ox-alpha context window pricing limits provider modalities Total results: 1 Ox Alpha - API Pricing & Providers https://openrouter.ai/stealth/ox-alpha ... and is free for a 1M-token context window, maximum output of 131072 tokens ... This limits the model's choice of tokens at ... (Wall: 3.14s | Timeout: 60s) tinyfish search query 'Ox Alpha TechCrunch OpenRouter anonymous third party August 2026 Zhiqu GLM 5.3' --pretty Output (28 earlier lines, showing 10 of 38) (ctrl+0 to expand) https://techcrunch.ai/articles/mystery-ai-ox-alpha-surfaces-on-openrouter-fueling-speculation-over-frontier-chinese-tech/ The mystery AI model, which recently surfaced on OpenRouter, features a 1.05 million-token context window and could be from China. Comment "FREE" and I'll send you the access. A frontier-tier ... https://www.instagram.com/reel/DcT2D9X-xJ/ The Chinese models are honest hard working Deasts fIVi+k US AI. (Wall: 2.18s | Timeout: 60s) Ox Alpha Emerges as Free Frontier AI Model on OpenRouter https://x.com/ifirending/2098042104603970 OpenRouter listed Ox Alpha from an anonymous provider, touting it for coding, Agentic tasks, and production work with text, image, ... (Wall: 2.08s | Timeout: 60s) TODO I. Identification 1/2 Identify Ox Alpha product and official listings Trace creator clues and model attribution II. Specifications 0/2 III. Synthesis 0/2 Searching deep reporting (esc) O GPT-5.A-Sol high tmp 0.27 -3% Research Ox Alpha and glm-5 Reviewing official usage limits and FAQ Read https://openrouter.ai/docs/api_reference/limits Requesting additional code lines 300 usage: number; // Number of credits used (all time) 301 usage_daily: number; // Number of credits used (current UTC day) 302 usage_weekly: number; // ... (current UTC week, starting Monday) 303 usage_monthly: number; // ... (current UTC Month) 304 305 306 BYDN_usage: number; // Same for external BYDN usage 307 BYDN_usage_daily: number; 308 BYDN_usage_weekly: number; 309 BYDN_usage_monthly: number; 310 311 is_free_tier: boolean; // Whether the user has paid for credits before 312 // rate_limit: ( ... ) // A deprecated object in the response, safe to ignore Investigating Ox Alpha usage limits tinyfish search query 'site:openrouter.ai "Ox Alpha" OR site:openrouter.ai "Ox Alpha" August 20 2026' --pretty Output Query: site:openrouter.ai "Ox Alpha" OR site:openrouter.ai "Ox Alpha" August 20 2026 Total results: 0 No results. (Wall: 4.02s | Timeout: 60s) Searching official announcement (esc) O GPT-5.A-Sol high tmp 0.55 -4% Research Ox Alpha and glm-5 Considering TinyFish fetch for dynamic JS tinyfish fetch content get --format markdown 'https://z.ai/blog/glm-5.3' Output {"results":[{"url":"https://z.ai/blog/glm-5.3","find_at":"","content":"..."}] Output (38 earlier lines, showing 10 of 38) (ctrl+0 to expand) ... changes to the NVIDIA wheels caused the Docker image build to fail, so we made ... -extra-index-url https://pypi.org/simple ... to restore successful builds. (Wall: 0.81s | Timeout: 90s) Reviewing GLM official was generation and pricing Read https://docs.z.ai/guides/overview/pricing TODO I. Identification 1/2 Identify Ox Alpha product and official listings Trace creator clues and model attribution II. Specifications 0/2 III. Synthesis 0/2 O GPT-5.A-Sol high tmp 0.77 -5% Research Ox Alpha and glm-5 Read (2) https://openrouter.ai/terms/stealth https://openrouter.ai/terms/stealth.md Identifying TinyFish content fetching tinyfish fetch content get --format markdown 'https://openrouter.ai/terms/stealth' | jq -r '.results[0].text' Output (95 earlier lines, showing 10 of 105) (ctrl+0 to expand) * (xii) use, or allow others to use, the Stealth Models in connection with any of the following: * (a) any illegal or unlawful activity, including activities involving violence or terrorism, exploitation or harm to children, ... * (b) military, warfare, nuclear industries or applications, espionage, or materials or activities subject to the International ... * (c) development of weapons, illegal drugs, or regulated/controlled substances; * (d) content intended to incite or promote violence, abuse, or bodily harm to any individual; * (e) creation of malicious code, malware, or computer viruses; or * (f) any action to intentionally circumvent or remove usage restrictions or other safety measures, or to enable functionality of ... Reviewing supplemental terms via TinyFish tinyfish fetch content get --format markdown 'https://openrouter.ai/terms/stealth/supplemental' | jq -r '.results[0].text' Output # Stealth Program Supplemental Terms This page lists all Stealth Models that have supplemental terms required by their respective Stealth Providers. By accessing a Stealth Model listed below, you agree to the applicable supplemental terms in addition to the Stealth Program End User License Agreement and the OpenRouter Terms of Service. If a Stealth Model is not listed below, no supplemental terms apply beyond the standard EULA. (Wall: 0.49s | Timeout: 90s) Planning dynamic model data extraction tinyfish agent run --url 'https://openrouter.ai/stealth/ox-alpha' --sync 'Inspect the complete Ox Alpha model page including Overview, Providers, activity, and any privacy/data policy scoops as tooltips. Return JSON exactly with fields: model_id, description, pricing, context_window, max_output, modalities, supported_parameters, provider_count. provider_names, provider_data_retention, provider_training_policy, moderation, availability_or_preview_notice, rate_limit_statement, warnings. Do not infer, use null for anything not shown.' Output Waiting for result... (Timeout: 300s) TODO I. Identification 1/2 Identify Ox Alpha product and official listings Trace creator clues and model attribution II. Specifications 0/2 III. Synthesis 0/2 Extracting dynamic model details (esc) O GPT-5.A-Sol high tmp 0.97 -7% Research Ox Alpha and glm-5 Flagging EULA contradictions and investigating user ID tracking Clarifying OpenRouter free model quota claims Read https://x.com/a-mysterious-ai-lab-16-offering-100-trillion-free-tokens-day-for-its-ox-alpha-model-as-evidence-points-to-zhipus-unreleased-glm-5/ Planning direct Reddit data retrieval Read https://www.reddit.com/r/singularity/comments/1ivfxxj/i_fingerprinted_ox_alpha_some_tokenizer_as_glm53/ Identifying Reddit JSON endpoint Read https://www.reddit.com/comments/1ovfxxj.json Testing TinyFish fetch functionality tinyfish fetch content get --format markdown 'https://www.reddit.com/r/singularity/comments/1ivfxxj/i_fingerprinted_ox_alpha_some_tokenizer_as_glm53/' | jq -r '.results[0].text' Output (283 earlier lines, showing 10 of 293) (ctrl+0 to expand) RevolutionalRedstone 3d ago THANK GOD china is here to ruin the American AI monopoly! Claudes and Codes have been sabotaging / dragging / lazy on purpose for a while now. The Chinese models are honest hard working Deasts fIVi+k US AI. (Wall: 19.17s | Timeout: 90s) tinyfish fetch content get --format markdown 'https://www.reddit.com/r/LocalLlama/comments/1vdlxk/ox_alpha_stealth_model_glm5_xir_mimo_v2_srv/' | jq -r '.results[0].text' Output Fetching wider tokenizer tests (esc) O GPT-5.A-Sol high tmp 1.73 -9% Research Ox Alpha and glm-5 Preparing bash curl API call curl -s -X POST https://opencode.ai/v1/chat/completions -H 'Content-Type: application/json' -d '{"model":"x-preview-f-free","messages":[{"role":"user","content":"hi"}],"top_p":0.8}' Output {"error":{"type":"server_error","message":"Error from provider (Console: Upstream request failed: [400] JSON parse error: Cannot deserialize value of type java.lang.Double from String \"hi\": not a valid Double value; nested exception is com.fasterxml.jackson.databind.exc.InvalidFormatException: Cannot deserialize value of type java.lang.Double from String \"hi\": not a valid Double value\n at [Source: (org.springframework.util.StreamUtils$NonClosingInputStream); line: 1, column: 78] (through reference chain: com.wds.api.box.sdk.v4.chat.ChatCompletionRequest[\"top_p\"])"}} Planning error reproduction with invalid role curl -s -X POST https://opencode.ai/v1/chat/completions -H 'Content-Type: application/json' -d '{"model":"x-preview-f-free","messages":[{"role":"bogus","content":"hi"}]}' Output {"error":{"type":"server_error","message":"Error from provider (Console: Upstream request failed: [422] Incorrect role information)"}} Researching Opencode API limits tinyfish search query 'site:opencode.ai ox alpha free one week 100 trillion tokens' --pretty Output Query: site:opencode.ai ox alpha free one week 100 trillion tokens Total results: 0 No results. (Wall: 4.10s | Timeout: 60s) Read https://x.com/hndxr/status/209854505783320429 Read https://x.com/OpenRouter/status/209954479923184269 Clarifying training restriction ambiguity tinyfish search query 'free for a week "Ox Alpha" Opencode' --pretty Output (28 earlier lines, showing 10 of 38) (ctrl+0 to expand) https://www.instagram.com/reel/DcRUMVT3M/ all free for a week. But who is actually behind Ox Alpha? Hashtags: #AI #Vibecoding #Opencode #OxAlpha #generativeAI. View 1 comment ... Somebody is handing 100 trillion tokens daily. Nobody knows who ... https://www.instagram.com/p/DcT2D9X-xJ/ A frontier AI model showed up yesterday with no name on it, and it's free for a week. It's called Ox Alpha. Opencode, the open source coding ... A mysterious stealth Ox Alpha model just came out of nowhere. It's free ... https://docs.z.ai/guides/capabilities/thinking_mode Thinking is activated by default in GLM-5.1 GLM-5.2 GLM-5.3 GLM-5.4 GLM-5.5 GLM-5.6 GLM-5.7 series, different from the default hybrid thinking in GLM-4.6. If you want to ... (Wall: 0.83s | Timeout: 60s) Read https://www.businessinsider.com/ox-alpha-ai-model-mystery-2024-8 Planning targeted GLM-5.3 search tinyfish search query 'site:z.ai OR site:docs.z.ai "Ox Alpha"' --pretty Output Query: site:z.ai OR site:docs.z.ai "Ox Alpha" Total results: 0 No results. (Wall: 3.48s | Timeout: 60s) TODO I. Identification 2/2 Identify Ox Alpha product and official listings Trace creator clues and model attribution II. Specifications 0/2 III. Synthesis 0/2 Checking flash readme evidence (esc) O GPT-5.A-Sol high tmp 2.92 -12% Research Ox Alpha and glm-5 Assessing OpenRouter usage and privacy Formulating Ox Alpha assessment and confidence labels Noting no separate media charges and context limits Ox Alpha: the short answer Ox Alpha is a temporary, free "Stealth" model exposed through OpenRouter under stealth/ox-alpha. OpenRouter says an anonymous third party develops and operates it. Best current attribution: [INFERENCE] Zhipu AI/2.ai is almost certainly operating the endpoint, and the underlying model is very likely GLM-5.3-derived. It is probably a previously unreleased multimodal GLM-5.3 core variant. Not the original GLM-5.3. Calling it "GLM-5.3 Flash" is premature. That name has not appeared in Z.ai's announcements, documentation, model catalog, or pricing table. The defensible description is: An unrevealed 2.ai-hosted GLM-5.3-class multimodal model-possibly a Flash, Air, V, Turbo, or Iotor derivative. Exact SKU unconfirmed. Confirmed specifications Source: OpenRouter's live model API (https://openrouter.ai/api/v1/models/stealth/ox-alpha/endpoint) and model page (https://openrouter.ai/stealth/ox-alpha). Property Confirmed value API model ID stealth/ox-alpha Added to OpenRouter August 20, 2026 Current price $0 input / $0 output Context window 1,048,576 tokens Maximum output 131,072 tokens Input modalities Text, images, video Output modal TODO III. Synthesis 2/2 Assess evidence confidence and unresolved claims Deliver sourced research with practical limits Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5 Images and video consume context after encoding; a "one-million-token window" does not mean one million tokens of text plus unlimited video. OpenRouter price Currently: Input: $0/M tokens Output: $0/M tokens No separate media fees No long-context surcharge shown Image/video inputs appear covered by the free preview rather than separately priced This is explicitly a preview. OpenRouter's Stealth Program EULA (https://openrouter.ai/terms/stealth) says stealth models are offered free for a limited period and can be removed without notice. There is no announced post-preview price. For comparison-not a prediction of Ox Alpha's eventual price-z.ai currently charges (https://docs.z.ai/guides/overview/pricing): Public Z.ai model Input / M Cached input / M Output / M GLM-5 $1.00 $0.20 $3.20 GLM-5.3 $1.40 $0.28 $4.40 GLM-5V-Turbo $1.20 $0.24 $4.00 Rate limits The Ox Alpha metadata currently reports: per_request_limits: null No published model-specific RPM or requests-per-day limit Only one upstream provider, so provider congestion cannot be routed to ano TODO III. Synthesis 2/2 Assess evidence confidence and unresolved claims Deliver sourced research with practical limits Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5 Who made it? Official answer Unknown. OpenRouter's announcement (https://x.com/openrouter/status/209854505783320429) says it is a new stealth model. The model page says: Developed and operated by a third-party provider who has chosen to remain anonymous during this preview. OpenRouter is the gateway, not the developer. Neither Z.ai nor another laboratory has officially claimed it. TechCrunch's current reporting (https://techcrunch.ai/articles/who-behind-the-stealth-model-ox-alpha/) reaches the same official conclusion. Technical answer: the endpoint points directly at Z.ai The strongest evidence is no longer output style or model self-identification. It is the serving infrastructure. I reproduced the malformed-request probe against Opencode's direct Ox Alpha route. Sending top_p: "abc" returned a Java deserialization trace containing: com.wds.paas.api.domain.v4.chat.ChatCompletionRequest["top_p"] That package and route structure correspond to Zhipu's Z.ai/api/paas/v4/chat/completions infrastructure. I also reproduced the invalid-role probe. Ox Alpha returned: [422] Incorrect role information That is the same distinctive error code/dialect reportd from Z.ai-hosted GLM endpoints. GLM weights hosted by another operator, such as DeepInfra, reportedly returned different validation errors. This identifies the serving operator, not merely tokenizer ancestry. The independently documented investigation is in Explain.ai's forensic report (https://www.explain.ai/blog/ox-alpha-what-we-know-mystery-ai-model-august-2026). Other evidence Tokenizer fingerprint Community tests reported 30/30 diverse probes matching GLM-5.3. A separate ob-string comparis Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5 Four controlled-video tests reportedly matched GLM-5V-Turbo's token accounting. Nitesh's model-level scoring, approximately 147 tokens/second scaling, and resolution scaling. Test points to a 94% multimodal stack, independently of the text tokenizer. Candidate elimination MLfN accepts audio; Ox Alpha does not. MLfN, Qwen and GLM-4.6 reportedly produced different tokenizer or encoder signatures. The early Microsoft MAI theory was based primarily on a weaker tokenizer classification and is now contradicted by the server trace and broader multilingual tests. Is it actually GLM-5? Probably GLM-family, but almost certainly not plain GLM-5 The original GLM-5 is an older public model. Ox Alpha more closely matches GLM-5.3's current contract. Ox Alpha GLM-5.3 Public GLM-5V-Turbo Feature Context 1,048,576 1M 200K Max output 131,072 128K 128K Text input Yes Yes Yes Image/video input Yes No-text only Yes Mandatory reasoning Yes Yes Not the primary identifying clue Efforts low/high/max low/high/max Multiple thinking modes Coding/agent focus Yes Yes Yes, including visual coding Operator fingerprint Z.ai-like Z.ai Ox Alpha effectively combines: GLM-5.3's 1M context, 128K output, sampling defaults, reasoning controls and coding behavior GLM-5V-Turbo's image/video encoding and multimodal coding stack That makes a new multimodal GLM-5.3-class derivative the most coherent explanation. Why the exact checkpoint remains unknown Tokenizer equality cannot distinguish GLM-5.2 from GLM-5.3 a Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5 It is GLM-5.3-rara rather than original GLM-5 High confidence It uses Z.ai's Multimodal / GLM-5V technology High confidence It is the exact public GLM-5.3 checkpoint Uncertain Its official name is "GLM-5.3 Flash" Low confidence / rumor It is plain GLM-5 Unlikely Is it actually good? There are strong early signals, but no trustworthy large-scale benchmark package yet. Reported community results include: 8/10 on one DeepSME sample 48/50 on a SWE-bench Verified Mini run 87.5% on Kingbench, versus a reported 91.25% for GLM-5.3 These are not comparable to audited, full benchmark runs: Ten tasks means one task shifts the result by ten percentage points. The 100-task suite covers a narrow subset of repositories. Old json-schema examination task. Agent scaffold, tool configuration and reasoning effort materially affect results. No official Ox Alpha benchmark card, trajectories, model report, or reproducible evaluation suite has been published. The safe conclusion: it appears to be a strong coding and agent model worth evaluating, not a demonstrated winner over Claude, GPT or GLM-5.3. Practical limitations 1. Temporary endpoint May disappear or be renamed without notice. 2. Full EULA & production dependency without a fallback model. 3. One provider No equivalent Ox Alpha provider fallback. Capacity errors can terminate requests. 4. Mandatory reasoning This feature must be disabled. Even 1M effort may add latency and token consumption. 4. Long context is a maximum, not a quality guarantee No published evidence establishes reliable recall or reason Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5 response_format is supported. OpenRouter does not advertise structured_outputs/JSON Schema support for this endpoint. Unknown internals No parameter count, active parameter count, architecture report, training data, license, safety card, or weights. No native web access It can call supplied tools, but the model is not inherently browsing the web. The data-policy problem There is a material contradiction. OpenRouter's launch thread says: The provider does not train on your prompts or completions. The current provider presentation also indicates unknown retention and no training. But the Stealth Program EULA (https://openrouter.ai/terms/stealth) says: says user content may be collected and shared with the anonymous provider, describes stealth access as consideration for training and improvement data, grants OpenRouter an irrevocable, perpetual license to process and sublicense submitted content to the stealth provider for training, evaluation, and improvement, sends personal data in prompts to that provider, and permits removal of the model without notice. The supplemental-terms page does not currently show an Ox-specific exception resolving this conflict. Operational decision: treat retention as unknown and training rights as unresolved. Do not send: private repositories, secrets or API keys, customer data, personal or regulated information, unreleased product plans, proprietary security findings. Use it for public code, synthetic tasks, disposable evaluations, and non-sensitive benchmarks until the provider is named and binding data terms are clear. Final verdict Ox Alpha is not just a random model branded by OpenRouter. The endpoint leaks Z.ai server implementation details, uses Z.ai's error dialect, matches the GLM tokenizer, matches GLM-5.3's reasoning and context contract, and appears to use GLM-5V-Turbo's video encoder. Most likely identity: [INFERENCE] an u Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.25 -13% Research Ox Alpha and glm-5

   Camera

   Static wide shot of a terminal window, with content scrolling vertically.

   Motion

   Rapid, continuous vertical scrolling of terminal output, simulating real-time agent activity. Text appears to be typed and executed.

3. 0:18 – 0:38

   Feature Teaser

   Accelerated Agent Workflow & Data Synthesis

   “(No spoken dialogue — High-energy audio transition with rising synth sweep and sub-bass drop/impact, leading into a pronounced beat with 808-style kicks and snares.)”

   On screen

   Unknown internals No parameter count, active parameter count, architecture report, training data, license, safety card, or weights. No native web access It can call supplied tools, but the model is not inherently browsing the web. The data-policy problem There is a material contradiction. OpenRouter's launch thread says: The provider does not train on your prompts or completions. The current provider presentation also indicates unknown retention and no training. But the Stealth Program EULA (https://openrouter.ai/terms/stealth) says: says user content may be collected and shared with the anonymous provider, describes stealth access as consideration for training and improvement data, grants OpenRouter an irrevocable, perpetual license to process and sublicense submitted content to the stealth provider for training, evaluation, and improvement, sends personal data in prompts to that provider, and permits removal of the model without notice. The supplemental-terms page does not currently show an Ox-specific exception resolving this conflict. Operational decision: treat retention as unknown and training rights as unresolved. Do not send: private repositories, secrets or API keys, customer data, personal or regulated information, unreleased product plans, proprietary security findings. Use it for public code, synthetic tasks, disposable evaluations, and non-sensitive benchmarks until the provider is named and binding data terms are clear. Final verdict Ox Alpha is not just a random model branded by OpenRouter. The endpoint leaks Z.ai server implementation details, uses Z.ai's error dialect, matches the GLM tokenizer, matches GLM-5.3's reasoning and context contract, and appears to use GLM-5V-Turbo's video encoder. Most likely identity: [INFERENCE] an u Reading model metadata definitions (esc) O GPT-5.A-Sol high tmp 3.47 -13% Research Ox Alpha and glm-5

   Camera

   Static wide shot of a terminal window, with content scrolling vertically.

   Motion

   Continued rapid vertical scrolling of terminal output, emphasizing speed and volume of information processing. The content is dense and technical.

4. 0:38 – 0:46

   Call to Action

   Brand Reinforcement & Call to Action

   “(No spoken dialogue — Subtle riser and impact at [0:38], maintaining the energetic beat.)”

   On screen

   TinyFish Give AI agents access to the entire Web www.tinyfish.ai

   Camera

   Static wide shot, centered logo and text overlay.

   Motion

   Logo and text fade-in with a subtle glow effect, returning to the initial pixel grid background.