Runway Gen-4.5 is a video-generation model with text-to-video and image-to-video modes. It can follow a brief that describes scene composition, subject action, camera choreography and a sequence of events. The useful result is a short generated shot that can be reviewed, retained or assembled with other footage.
Text-to-video starts from a written scene description. Include both the visible content and the motion: subject, setting, lighting, framing, action and camera behavior. This mode is useful for a new scene, B-roll, background plate or visual concept where the starting frame does not have to reproduce a supplied image.
Image-to-video starts from an image that acts as the first frame. That image supplies composition, subject matter, lighting and style. Focus the written prompt on what should happen next: the subject's action, movement in the environment, camera motion and the progression of the shot.
Prepare a clear image with the important details already visible. Blurred faces, hands or other artifacts can become more noticeable when animated. A reference that suggests rapid motion through dust or motion blur can also conflict with a prompt that asks for a stationary subject.
Use clear natural language and positive descriptions of the intended result. A text-to-video brief can state the camera view, subject and action first, then add the scene treatment. A detailed description is useful when the shot has defined requirements; a simpler one gives more room for exploration. There is no ideal word count that replaces clarity.
For example, describe a close view of a brass desk lamp as its light turns on, followed by a slow camera move that reveals a notebook beside it. State the visible materials and light for text-to-video. If a supplied image already establishes those details, the image-to-video prompt can concentrate on the light change and camera reveal.
Avoid contradictory directions. A locked-off view with subtle subject motion is different from an energetic handheld tracking shot. Add the details that define the useful shot rather than a long list of unrelated visual styles.
The maker's camera guide distinguishes framing, angles, movement and focus. A close-up changes what fills the frame, a low angle changes perspective, a pan rotates the camera's view and a tracking move follows a subject. A rack-focus instruction shifts the focus between visible subjects. State what should be in view or revealed as the movement develops.
Sequential prompts can express events in ordinary language or with rough timestamps. Those timestamps guide the order and approximate timing; they are not a guarantee of frame-exact execution. Give the requested actions enough time to occur naturally within the selected clip duration.
For a static shot, describe the motion that belongs inside the frame and reinforce the intended still camera. The maker notes that completely motionless views can be difficult, particularly for some wide or establishing shots. Review small unwanted movements in the actual result rather than assuming a static-camera phrase makes them impossible.
Gen-4.5 supports clip durations from 2 to 10 seconds. A longer choice can accommodate several ordered actions, while a brief reaction or reveal can use a shorter shot. Its documented output is 720p, with 24 or 25 frames per second in the maker's generation workflow.
The text-to-video specification lists 16:9 at 1280 × 720 pixels. Image-to-video additionally lists 9:16 at 720 × 1280, 1:1 at 960 × 960, 4:3 at 1104 × 832, 3:4 at 832 × 1104 and 21:9 at 1584 × 672. Keep the mode and actual dimensions together rather than assuming both input modes offer the same aspect-ratio set.
An image's ratio can establish the default composition, and changing it can crop the input. Check the subject's placement before judging motion: a correct action is not useful if a key part of the source image has been cut out.
For image-led work, the maker describes continuing a sequence by using a completed clip's last frame as the image input to another generation. The clips are then combined in a video editor, with their shared frame and timing handled in the edit. This is a sequence of separate shots, rather than one unlimited-duration generation.
Retain the end frame and the corresponding brief so that the next shot starts from the intended visual state. Inspect continuity in the joined footage, including subject appearance, camera direction and the transition between clips.
After generation, review the full clip and retain the completed output. The producer's workflow also provides separate upscaling and production-export choices such as ProRes or a PNG sequence. Those are additional workflow options; the native generation dimensions remain distinct from a later upscaled result.
Use Gen-4.5 for cinematic concepts, animated visual ideas, B-roll and short scene development. Keep the prompt, input image where used, duration, mode and returned clip together. Check action order, camera behavior, subject details, unwanted cuts and the intended framing, then make a focused correction to the shot.
", "aboutText": "Runway Gen-4.5 is a video-generation model with text-to-video and image-to-video modes. It can follow a brief that describes scene composition, subject action, camera choreography and a sequence of events. The useful result is a short generated shot that can be reviewed, retained or assembled with other footage.\n\n### Choose the input mode for the shot\n\nText-to-video starts from a written scene description. Include both the visible content and the motion: subject, setting, lighting, framing, action and camera behavior. This mode is useful for a new scene, B-roll, background plate or visual concept where the starting frame does not have to reproduce a supplied image.\n\nImage-to-video starts from an image that acts as the first frame. That image supplies composition, subject matter, lighting and style. Focus the written prompt on what should happen next: the subject's action, movement in the environment, camera motion and the progression of the shot.\n\nPrepare a clear image with the important details already visible. Blurred faces, hands or other artifacts can become more noticeable when animated. A reference that suggests rapid motion through dust or motion blur can also conflict with a prompt that asks for a stationary subject.\n\n### Describe visuals and motion directly\n\nUse clear natural language and positive descriptions of the intended result. A text-to-video brief can state the camera view, subject and action first, then add the scene treatment. A detailed description is useful when the shot has defined requirements; a simpler one gives more room for exploration. There is no ideal word count that replaces clarity.\n\nFor example, describe a close view of a brass desk lamp as its light turns on, followed by a slow camera move that reveals a notebook beside it. State the visible materials and light for text-to-video. If a supplied image already establishes those details, the image-to-video prompt can concentrate on the light change and camera reveal.\n\nAvoid contradictory directions. A locked-off view with subtle subject motion is different from an energetic handheld tracking shot. Add the details that define the useful shot rather than a long list of unrelated visual styles.\n\n### Camera terms and sequenced action\n\nThe maker's camera guide distinguishes framing, angles, movement and focus. A close-up changes what fills the frame, a low angle changes perspective, a pan rotates the camera's view and a tracking move follows a subject. A rack-focus instruction shifts the focus between visible subjects. State what should be in view or revealed as the movement develops.\n\nSequential prompts can express events in ordinary language or with rough timestamps. Those timestamps guide the order and approximate timing; they are not a guarantee of frame-exact execution. Give the requested actions enough time to occur naturally within the selected clip duration.\n\nFor a static shot, describe the motion that belongs inside the frame and reinforce the intended still camera. The maker notes that completely motionless views can be difficult, particularly for some wide or establishing shots. Review small unwanted movements in the actual result rather than assuming a static-camera phrase makes them impossible.\n\n### Duration, dimensions and frame rate\n\nGen-4.5 supports clip durations from 2 to 10 seconds. A longer choice can accommodate several ordered actions, while a brief reaction or reveal can use a shorter shot. Its documented output is 720p, with 24 or 25 frames per second in the maker's generation workflow.\n\nThe text-to-video specification lists 16:9 at 1280 × 720 pixels. Image-to-video additionally lists 9:16 at 720 × 1280, 1:1 at 960 × 960, 4:3 at 1104 × 832, 3:4 at 832 × 1104 and 21:9 at 1584 × 672. Keep the mode and actual dimensions together rather than assuming both input modes offer the same aspect-ratio set.\n\nAn image's ratio can establish the default composition, and changing it can crop the input. Check the subject's placement before judging motion: a correct action is not useful if a key part of the source image has been cut out.\n\n### Build a longer sequence from completed shots\n\nFor image-led work, the maker describes continuing a sequence by using a completed clip's last frame as the image input to another generation. The clips are then combined in a video editor, with their shared frame and timing handled in the edit. This is a sequence of separate shots, rather than one unlimited-duration generation.\n\nRetain the end frame and the corresponding brief so that the next shot starts from the intended visual state. Inspect continuity in the joined footage, including subject appearance, camera direction and the transition between clips.\n\n### Retain and review the production artifact\n\nAfter generation, review the full clip and retain the completed output. The producer's workflow also provides separate upscaling and production-export choices such as ProRes or a PNG sequence. Those are additional workflow options; the native generation dimensions remain distinct from a later upscaled result.\n\nUse Gen-4.5 for cinematic concepts, animated visual ideas, B-roll and short scene development. Keep the prompt, input image where used, duration, mode and returned clip together. Check action order, camera behavior, subject details, unwanted cuts and the intended framing, then make a focused correction to the shot.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "e35a53d3d7b02d937c3d4b3db688dc908e2f8a037b64cadd434f8d068bcca064", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "runway/gen-4.5Runway Gen-4.5 creates videos from a scene description or an image with motion direction. Its 2–10-second clips, explicit camera prompting and separate text-to-video and image-to-video composition choices support cinematic concepts, B-roll and animated visual scenes.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
Runway Gen-4.5 is a video-generation model with text-to-video and image-to-video modes. It can follow a brief that describes scene composition, subject action, camera choreography and a sequence of events. The useful result is a short generated shot that can be reviewed, retained or assembled with other footage.
Text-to-video starts from a written scene description. Include both the visible content and the motion: subject, setting, lighting, framing, action and camera behavior. This mode is useful for a new scene, B-roll, background plate or visual concept where the starting frame does not have to reproduce a supplied image.
Image-to-video starts from an image that acts as the first frame. That image supplies composition, subject matter, lighting and style. Focus the written prompt on what should happen next: the subject's action, movement in the environment, camera motion and the progression of the shot.
Prepare a clear image with the important details already visible. Blurred faces, hands or other artifacts can become more noticeable when animated. A reference that suggests rapid motion through dust or motion blur can also conflict with a prompt that asks for a stationary subject.
Use clear natural language and positive descriptions of the intended result. A text-to-video brief can state the camera view, subject and action first, then add the scene treatment. A detailed description is useful when the shot has defined requirements; a simpler one gives more room for exploration. There is no ideal word count that replaces clarity.
For example, describe a close view of a brass desk lamp as its light turns on, followed by a slow camera move that reveals a notebook beside it. State the visible materials and light for text-to-video. If a supplied image already establishes those details, the image-to-video prompt can concentrate on the light change and camera reveal.
Avoid contradictory directions. A locked-off view with subtle subject motion is different from an energetic handheld tracking shot. Add the details that define the useful shot rather than a long list of unrelated visual styles.
The maker's camera guide distinguishes framing, angles, movement and focus. A close-up changes what fills the frame, a low angle changes perspective, a pan rotates the camera's view and a tracking move follows a subject. A rack-focus instruction shifts the focus between visible subjects. State what should be in view or revealed as the movement develops.
Sequential prompts can express events in ordinary language or with rough timestamps. Those timestamps guide the order and approximate timing; they are not a guarantee of frame-exact execution. Give the requested actions enough time to occur naturally within the selected clip duration.
For a static shot, describe the motion that belongs inside the frame and reinforce the intended still camera. The maker notes that completely motionless views can be difficult, particularly for some wide or establishing shots. Review small unwanted movements in the actual result rather than assuming a static-camera phrase makes them impossible.
Gen-4.5 supports clip durations from 2 to 10 seconds. A longer choice can accommodate several ordered actions, while a brief reaction or reveal can use a shorter shot. Its documented output is 720p, with 24 or 25 frames per second in the maker's generation workflow.
The text-to-video specification lists 16:9 at 1280 × 720 pixels. Image-to-video additionally lists 9:16 at 720 × 1280, 1:1 at 960 × 960, 4:3 at 1104 × 832, 3:4 at 832 × 1104 and 21:9 at 1584 × 672. Keep the mode and actual dimensions together rather than assuming both input modes offer the same aspect-ratio set.
An image's ratio can establish the default composition, and changing it can crop the input. Check the subject's placement before judging motion: a correct action is not useful if a key part of the source image has been cut out.
For image-led work, the maker describes continuing a sequence by using a completed clip's last frame as the image input to another generation. The clips are then combined in a video editor, with their shared frame and timing handled in the edit. This is a sequence of separate shots, rather than one unlimited-duration generation.
Retain the end frame and the corresponding brief so that the next shot starts from the intended visual state. Inspect continuity in the joined footage, including subject appearance, camera direction and the transition between clips.
After generation, review the full clip and retain the completed output. The producer's workflow also provides separate upscaling and production-export choices such as ProRes or a PNG sequence. Those are additional workflow options; the native generation dimensions remain distinct from a later upscaled result.
Use Gen-4.5 for cinematic concepts, animated visual ideas, B-roll and short scene development. Keep the prompt, input image where used, duration, mode and returned clip together. Check action order, camera behavior, subject details, unwanted cuts and the intended framing, then make a focused correction to the shot.
POST /v1/videostext, imagevideo| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits runway/gen-4.5 | Default— |
Fieldprompt | Typestring | RequirementRequired | Values & limits | Default— |
Fieldduration | Typeinteger | RequirementRequired | Values & limits 2, 3, 4, 5, 6, 7, 8, 9, 10 Minimum: 1 Maximum: 600 | Default— |
Fieldresolution | Typestring | RequirementOptional | Values & limits 720p | Default— |
Fieldaspect_ratio | Typestring | RequirementOptional | Values & limits 16:9, 9:16 | Default— |
Fieldgenerate_audio | Typeboolean | RequirementOptional | Values & limits false | Default— |
Fieldframe_images | Typearray | RequirementOptional | Values & limits Minimum items: 1 Maximum items: 1 Input content: Array of frame objects. Each object has type=image_url, image_url containing url, and frame_type. Use an HTTPS image URL and set frame_type to first_frame. Each frame role appears at most once. Exact keys: type, image_url, frame_type Conditions: These keys describe each frame object; image_url contains only url. Omit frame_images for a text-only request. | Default— |
Fieldinput_references | Typearray | RequirementOptional | Values & limits | Default— |
Fieldsize | Typestring | RequirementOptional | Values & limits 1280x720, 720x1280 Pattern: ^[1-9][0-9]*x[1-9][0-9]*$ | Default— |
Fieldseed | Typeinteger | RequirementOptional | Values & limits Minimum: -9,223,372,036,854,775,808 Maximum: 9,223,372,036,854,775,807 | Default— |
Fieldcallback_url | Typestring | RequirementOptional | Values & limits Pattern: ^https:// | Default— |
Fieldsession_id | Typestring | RequirementOptional | Values & limits | Default— |
Fieldtrace | Typeobject | RequirementOptional | Values & limits | Default— |
Fielduser | Typestring | RequirementOptional | Values & limits | Default— |
Text to video · A written prompt. · Required: model, prompt, duration
Provide prompt as text and duration as whole seconds. Choose a supported resolution; add an aspect ratio or use a listed size.
model: Type: string · runway/gen-4.5
prompt: Type: string · Maximum UTF-8 bytes: 10,000 · Non-whitespace required: Yes
Image to video · A written prompt and a first-frame image. · Required: model, prompt, duration, frame_images
frame_images is an array. Each frame object has type, image_url and frame_type; image_url contains url. Set type to image_url. Use first_frame. Each frame role appears at most once. Supply a public HTTPS image URL for each frame. Choose a supported resolution; add an aspect ratio or use a listed size.
Exact keys: type, image_url, frame_type
model: Type: string · runway/gen-4.5
prompt: Type: string · Maximum UTF-8 bytes: 10,000 · Non-whitespace required: Yes
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, prompt, duration |
Parametermaterial_types | Contracttext, image |
Parameteradditional_properties | Contractfalse |
Parameteroptional_controls | Contractresolution, aspect_ratio, generate_audio, frame_images, input_references, size, seed, callback_url, session_id, trace, user |
Lifecycle: submit_once_poll_original
GET /v1/videos/{public_id}
GET /v1/videos/{public_id}/content
The displayed rates estimate the cost before generation. The final charge is calculated from the cost recorded for the original generation. Measured output duration and forecast quantities are not the final billable quantity.
Tariff references do not add request options. Supported parameters are listed for the selected method.
Match each charge to the conditions shown. Use one rate for each set of alternative configurations.
| Configuration | Rate schedule | Forecast quantity | Billing unit | Billing conditions | Charge |
|---|---|---|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Forecast quantity— | Billing unit— | Billing conditions— | Charge Input · Input tokens · Estimate for comparison only — / 1K tokens Output · Output tokens · Estimate for comparison only — / 1K tokens Input · Cache read · Estimate for comparison only — / 1K tokens Input · Cache write · Estimate for comparison only — / 1K tokens Input · Cache write, 1 hour · Estimate for comparison only — / 1K tokens |
| ConfigurationVideo price | Rate schedule— | Forecast quantity— | Billing unitper video second | Billing conditionsNo additional condition is declared in this pricing key. | Charge Output · Video price · Estimate for comparison only $0.24 per video second |
| ConfigurationEstimated generation cost | Rate schedule— | Forecast quantityone generation estimate | Billing unit— | Billing conditions— | Charge Output · Estimated generation cost · Estimate for comparison only $0.48 estimate before generation This estimate is not an additional charge or a fixed rate for the completed generation. |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
Model-level reference schedules do not change method selection, quotes or charges. Active charges use the selected Ailoot method rates.
| Configuration | Billing conditions | Charge |
|---|---|---|
| ConfigurationVideo price | Billing conditionsNo additional condition is declared in this pricing key. | Charge Output · Video price $0.24 per video second Published price reference for the configuration shown. Final charges follow the actual billed configuration. |
set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\npython3 - <<'PY_REQUEST'\nfrom pathlib import Path\nPath('request.json').write_text('{"aspect_ratio":"16:9","duration":8,"model":"runway/gen-4.5","prompt":"A serene mountain landscape at sunset","resolution":"720p"}', encoding='utf-8')\nPY_REQUEST\ncurl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/videos \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \\\n --data-binary @request.json -D response-headers.txt -o result.json\npython3 - <<'PY_RESULT'\nimport json\nfrom pathlib import Path\nr=json.loads(Path('result.json').read_text())\nprint(json.dumps(r,ensure_ascii=False,indent=2))\nprint('Keep the complete JSON and response headers; do not resubmit an unknown creation.')\nPY_RESULT\nsigned owned content URLs
GET /v1/videos/{public_id}GET /v1/videos/{public_id}/contentset -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\npython3 - <<'PY_STATUS_URL'\nimport json\nfrom pathlib import Path\nfrom urllib.parse import quote\nr=json.loads(Path('result.json').read_text())\nif not isinstance(r.get('id'),str) or not r['id']:raise SystemExit('Keep the original response and headers for reconciliation; no returned ID')\nPath('status-url.txt').write_text('https://api.ailoot.ai/v1/videos/'+quote(r['id'],safe=''))\nPY_STATUS_URL\ncurl --fail-with-body --silent --show-error "$(cat status-url.txt)" \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -D status-headers.txt -o status.json\npython3 - <<'PY_RESULT'\nimport json\nfrom pathlib import Path\nr=json.loads(Path('status.json').read_text())\nprint(json.dumps(r,ensure_ascii=False,indent=2))\nprint('Keep the complete JSON and response headers; do not resubmit an unknown creation.')\nPY_RESULT\nset -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\nexport AILOOT_API_KEY\npython3 - <<'PY_ASSETS'\nimport json,os,subprocess\nfrom pathlib import Path\nfrom urllib.parse import urljoin,urlsplit\nr=json.loads(Path('status.json').read_text())\nif r.get('status')!='completed':raise SystemExit('Keep the original ID and read its status; never recreate an unknown job')\nif not isinstance(r.get('data'),list) or not r['data']:raise SystemExit('Keep the complete original status; no completed outputs supplied')\nurls=[]\nfor i,u in enumerate(r.get('data',[])):\n if not isinstance(u,str):raise SystemExit('Keep the original response; unexpected asset shape')\n urls.append((str(i),u))\nfor label,path in urls:\n u=urljoin('https://api.ailoot.ai',path);z=urlsplit(u)\n if z.scheme!='https' or z.netloc!='api.ailoot.ai' or z.username or z.password:raise SystemExit('Use the actual returned Ailoot owned content URL')\n target=Path('owned-'+label+'.bin')\n subprocess.run(['curl','--fail-with-body','--silent','--show-error',u,'-H','Authorization: Bearer '+os.environ['AILOOT_API_KEY'],'-D',str(target)+'.headers','-o',str(target)],check=True)\nprint('All returned owned URLs saved; complete metadata, timing and usage remain in the JSON.')\nPY_ASSETS\nSave every completed video or returned file; retrieve every actual returned asset URL.
Keep every actual owned output and its returned measured metadata.
503 result_pending_check + originalid + X-Should-Retry:false; never replay original
| Field | Type | Meaning |
|---|---|---|
Fieldindex | Typeinteger | MeaningZero-based position of this video in the outputs list. |
Fieldcontent_url | Typestring | MeaningSigned HTTPS download URL for this completed video. |
Fieldsha256 | Typestring | null | MeaningSHA-256 digest of the original video bytes; null if unavailable. |
Fieldsize_bytes | Typeinteger | null | MeaningSize of the original video file in bytes; null if unavailable. |
Fieldduration_ms | Typeinteger | null | MeaningMeasured video duration in milliseconds; null if unavailable. |
Fieldwidth | Typeinteger | null | MeaningVideo width in pixels; null if unavailable. |
Fieldheight | Typeinteger | null | MeaningVideo height in pixels; null if unavailable. |
Fieldvideo_codec | Typestring | null | MeaningVideo codec identifier; null if not recorded. |
Fieldhas_audio | Typeboolean | null | MeaningWhether the video contains an audio track; null if not measured. |
Fieldaudio_codec | Typestring | null | MeaningAudio codec identifier; null when no identifier is recorded. |
| Response | Condition | Next step |
|---|---|---|
Response404 · not_found_error | ConditionVideo access is unavailable, or this video ID was not found. | Next stepCheck video access. For an existing video, use its original public ID and the same API key. |
Response401 · authentication_error | ConditionThe API key does not match the authenticated account. | Next stepCheck your API key and account access before submitting. |
Response400 · invalid_request_error | ConditionThe request could not be read, or the download index is invalid. | Next stepCorrect the request body. For a download, choose an output index returned by the original task. |
Response422 · price_unavailable | ConditionThe selected video configuration cannot be priced. | Next stepChoose a supported video configuration and complete its required options before submitting a corrected request. |
Response422 · invalid_request_error | ConditionA video input or original continuation task is invalid. | Next stepUse valid HTTPS inputs. For continuation, use your completed original public video ID with the same API key. |
Response402 · billing_error | ConditionThe available balance or limits do not allow a reservation. | Next stepCheck your available balance and limits before sending a new request. |
Response503 · server_error | ConditionVideo verification, storage or service access is temporarily unavailable. | Next stepWait for service availability. If you already have a video ID, check that original task instead of creating another generation. |
Response503 · result_pending_check | ConditionThe original video result and charge need checking. | Next stepKeep the returned video ID and check that task. Do not send a replacement generation request while it is being checked. |
Response503 · owned_result_unavailable | ConditionThe existing video download is temporarily unavailable. | Next stepCheck the same video task later and download its existing output. A new generation creates a separate operation. |
Response409 · video_not_ready | ConditionThe video content has not been published. | Next stepCheck the original task status. Download after it reports completed; if it reports failed, follow its charge-check notice. |
Ailoot Ultra runway/gen-4.5
Use the method paths shown above.